Research article

PARVUS miniaturised robot for improved exibility in micro production

rgen Hesselbach and Reinhard Degen Arne Burisch, Jan Wrege, Annika Raatz, Ju
Institute of Machine Tools and Production Technology (IWF), Technical University of Braunschweig, Braunschweig, Germany
Abstract Purpose Until now, the size range of most machines for precision assembly was much larger than the size of the pieces to be handled or the necessary workspace. Flexibly scalable miniaturised production machines can help to develop much more exible micro production systems. The paper aims to describe the development of a micro-parallel-SCARA robot adapted in size to MEMS products. Design/methodology/approach The robot consists of a miniaturised parallel structure, which provides a high level of accuracy in a workspace of 60 45 20 mm3. It has a base area of 130 170 mm2 and offers four degrees of freedom. Findings Based on simulations, the degree of miniaturisation in terms of a smaller structure and a high level of accuracy is determined. The results show that a miniaturised hybrid robot with a plane parallel structure driven by miniaturised zero-backlash gears and electric motors can reach a theoretical repeatability better than 1 mm. Research limitations/implications The rst prototype provides good prospects that the concept will be used in a visionary desktop-factory. As regards the accuracy parameters of the robot, there will be further efforts to optimise the robots structure and drive mechanism. Practical implications The repeatability of this rst prototype is better than 14 mm. A better stiffness of optimised micro-gears and joints of the structure will guarantee a much better repeatability. Originality/value The paper illustrates that the Parvus is one of the smallest industrial robots for micro assembly equipped with a full range of functionalities like conventional industrial robots. Keywords Assembly, Robotics, Microcontrollers Paper type Technical paper

1. Introduction
Today, the trend of miniaturisation with regard to product development is leading to an increasing gap between the dimensions of the products and the production systems used. A lot of micro-products are assembled by hand which causes high assembly costs that amount to 20 to 80 per cent of the total production costs (Koelemeijer and Jacot, 1999). Assembly lines for millimetre-sized products often measure some tens of meters and are mostly too expensive for smalland medium-sized businesses. In recent years, the miniaturisation of production systems has been widely discussed in research papers. Potential for development can be found in the economic and technologic benets of this strategy. On the one hand, it is assumed that small dimensions offer the possibility for high modular system designs, as well as improved scalability and exibility in the manufacturing base of assembly lines. On the other hand, it is assumed that small footprints, low power consumption, minor amounts of maintenance and initial costs promise better cost
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0144-5154.htm

Assembly Automation 27/1 (2007) 65 73 q Emerald Group Publishing Limited [ISSN 0144-5154] [DOI 10.1108/01445150710724720]

effectiveness. Already in 1990, a research group of the Mechanical Engineering Laboratory (MEL) in Japan (Tsukuba) estimated that a 1/10 size-reduction of production machines could lead to a decrease of energy consumption of about 1/100 compared to a conventional factory (Okazaki et al., 2002). Nowadays the MEL with its Desktop Machining Factory is deemed to be a pioneer in the eld of micro factory (Okazaki et al., 2002; Tanaka, 2001). Basic ideas and questions related to desktop factories are also discussed by Breguet and Bergander (2001) in his paper Toward the Personal Factory. Deliberating about advantages and disadvantages of desktop factories, he describes the vision of conventional factories and desktop factories coexisting in future times. The aim should not be an all-purpose desktop factory but a high modular system adapted in size to MEMS as presented in the concepts of Gaugel et al. (2003) or Rochdi et al. (2003). These concepts combine individual assembly modules on a product-neutral platform used for the feeding of parts to individual process stations. Furthermore, industrial producers such as Klocke Nanotechnik (Dr Volker Klocke Nanotechnik, Pascalstr. 17, 52076 Aachen, Germany, available at: www. nanomotor.de) or MiLaSys technologies GmbH (RobertBosch-Str. 3, 71088 Holzgerlingen, Germany, available at: www.milasys.de) are making rst efforts to sell miniaturised micro production systems. 65

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

2. Potentials for miniaturised robots in micro production systems

The objective of most of the previously developed concepts is the general improvement of exibility and cost effectiveness of micro assembly and desktop factories. Although no desktop factories actually exist in industrial production, which could prove these assumptions, industry and research push the development of cheaper and smaller production technology. The gap between piezo driven, small walking micro robots and conventional robots has to be lled by cost-efcient miniaturised handling systems. A possible solution is to nd out the highest degree of miniaturisation of conventional robot technology, using innovative miniaturised machine parts. The feedback of industrial partners and the variety of possible applications showed that the improvement of exibility by utilising such miniaturised production machines could be important for the overall production process. In the following, the potentials of exibility of miniaturised production machines will be discussed. First of all, exibility can be subdivided into exibility of function exibility of piece number exibility of location and exibility of property. 2.1 Flexibility of function Nowadays, compared to conventional industrial robots, miniaturised robots can be equipped with a similar range of functionalities (e.g. degrees of freedom (DOF), sensors and tools). The combination of such miniaturised machines with freely programmable control systems, miniaturised drive systems and micro grippers can lead to the development of a micro production system with a high range of functionalities and motion-sequences. 2.2 Flexibility of production volume Less required space of miniaturised production machines makes it possible to replace a conventional robot by several miniaturised robots with increased productivity with regard to the space. These concepts, known as the desktop factory concepts, increase the availability and the production volume. Hence, the manufacturer is more exible in the number and variety of produced pieces. 2.3 Flexibility of placement Miniaturised production systems with a high density of functionality are exible in location. They can be placed in conventional clean rooms as well as in local clean room cells. Furthermore, companies prot from a high exibility of placement with regard to the development of processes as well as the expansion and relocation of the company. 2.4 Flexibility of property Most conventional production systems are not sufcient for future applications. For most companies, a complete substitution of these production systems is impossible, because the newly acquired expensive systems have to amortise. Thanks to miniaturised production machines, already existing bigger production systems can be equipped with extended functionalities. Owing to the smaller amount of space, miniaturised robots can be integrated into bigger machines. The symbiosis of conventional technology and micro technology leads to improved exible properties. Based on the aforementioned assumptions, most of the current problems in micro assembly or the semiconductor 66

industry can be solved by integrating a miniaturised robot in two ways (Figure 1): 1 as a component for miniaturised production systems such as visionary desktop factories, e.g. in micro production or micro assembly industry; and 2 as a miniaturised production machine integrated into a conventional bigger machine, e.g. in testing machines for conductor boards. The challenge of the Parvus Project is to develop a miniaturised precision industrial robot with the full functional range of larger models. Independent of its future application area, however, dependent on the aforementioned way (1) or (2) a miniaturised robot is integrated into the system, some basic requirements had to be determined. The main questions were the required size of the workspace and the desired performance specications regarding accuracy and velocity. The latter were determined by the presented design studies and the design constraints, especially the available size of system components such as servo drives, gearboxes or encoders and switches. Indications of a suitable workspace were provided by the size of innovative micro products of micro sensor electronic, telecommunication or biomedical science. These hybrid products consist of several metal, glass, plastic or ceramic components, e.g. electronic components, micro lenses, glass bres, photodiodes and prisms of a few millimetres in size. These components need a high-precision pick-and-place assembly process with a positioning accuracy up to 1 mm or better, whereas the overall sizes of the whole hybrid microproducts do not exceed the size of a chip card (Figure 2).

3. State-of-the-art of precision robots

A survey that categorises precision robots is shown in Figure 3. The simplest classication is into serial, parallel and hybrid structures, which in turn can be subdivided into further categories. The rst category covers Cartesian robots. These are typically very large in comparison to the components to be handled and are often, as a result, very expensive. However, they do provide repeatability between 1 and 3 mm, as demonstrated by, for example, the Sysmelec Autoplace 411. The second category covers serial SCARA robots, which have a large workspace in relation to their physical size, but only achieve a repeatability of 10 mm, even in the case of the most accurate designs. In the eld of parallel robots, there are not many examples in industrial use. The Mitsubishi RP-X, as a commercial parallel robot, is an exception and achieves a repeatability of 5 mm. Most other developments in this area are limited to university research projects, in particular at the Technical University of Braunschweig in Germany, where extensive experience with parallel structures has been gathered, for example, with the Triglide robot, which has achieved repeatability better than 1 mm (Hesselbach et al., 2004a, b, c). Further examples for miniaturised robots in research are the microdelta (CSEM, Centre Suisse dElectronique et de Microtechnique SA, Switzerland, available at: www.csem.ch) and the miniaturised Cartesian robot from AAL (Assembly Automation Laboratory, Tampere University of Technology, Institute of Production Engineering, Finland, available at: http://pe.tut./aal/).

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Figure 1 Two ways of system integration for the Parvus

Figure 2 Concept of micro laboratory on a chip card, thin XXS Microtechnology AG

Figure 3 Survey of high precision assembly robots

67

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Nowadays the existing high precision robots are relatively large and expensive. However, a growing market demanding for smaller and cheaper robotic devices for precision positioning and assembly is coming up. The minimization of conventional industrial robots is even in progress in commercial products of Yamaha or Mitsubishi. These robots reach a repeatability of 5 mm. A further miniaturisation of such industrial robots is possible because of new enabling technologies, in particular zero-backlash micro-gears and highly dynamic micro-motors with integrated incremental encoders, which allow proven robot arm structures to be miniaturised. These scalable miniaturised structures lead to improved dynamic properties and process speed because of their reduced dynamic mass. Furthermore, they allow the use of proven control technology and avoid the complexity of alternative actuator technologies such as piezo actuators (Hesselbach et al., 2004b, c).

5. Robot design approaches

At the beginning of the design phase, some basic specications were determined and quantitative specications were xed (Table I). The aim was to nd a simple modular structure with a small envelope and easy access to the working area of the robot. The robot should provide four DOF for a typical pick-and-place assembly process with the orthogonal x- and y-axes for positioning an assembly part in the xy-plane, the orthogonal z-axis for joining it and the rotational C-axis for its orientation. A pre-selection for a possible kinematics structure of the robot was identied taking a closer look on conventional industrial robots. In this principle design phase, two different plane kinematic structures were compared: the serial SCARA structure, e.g. used in the Yamaha YK 120, and the parallel SCARA structure known from the Mitsubishi Melfa series. The decision whether serial or parallel structure should be used was supported by a detailed positioning sensitivity analysis at all points in the workspace of the conceptual designs. The kinematic transfer functions according the robot drives q and the world coordinates r are basically given by the direct kinematic problem and the inverse kinematic problem of both structures: q q1 ; . . . ; qF T ; F DOF _

4. Innovative drive systems/components

In recent years, commercially miniaturised electric motors and high-resolution encoders have been coming up. In combination with micro-gears, as developed by Micromotion GmbH (An der Fahrt 13, 55124 Mainz, Germany, available at: www.micromotion-gmbh.de), these components can serve as adequate drives for micro assembly systems. The Micro Harmonic Drivew is the only micro-gear currently available that offers the same positioning accuracy as the large-scale Harmonic Drive gears used in industrial robots. Figure 4 shows the basic design of this gear, which uses only six components to achieve reduction ratios between 160:1 and 1000:1. These ratios are necessary to create adequate torques from currently available micro-motors, which are capable of rotational speeds up to 100,000 rpm with torques of only a few m nm. Related to the LIGA-technique, the single gear wheels of the Micro Harmonic Drivew are manufactured by electroplating and consist of a nickel-iron-alloy. Owing to the high yield point of 1,500 N/mm2, the low elastic modulus of 165,000 N/mm2 and its good fatigue endurance, this electroplated alloy possesses the necessary properties for perfect functioning of the exible gear wheels of this micro gear system. By providing an angular repeatability of ten arc seconds, the Micro Harmonic Drivew gear is the only micro-gear currently available that provides sufcient accuracy for a micro-robot meeting our requirements, given in the next section. Figure 4 Micro Harmonic Drivew gear components (Micromotion GmbH)

r X TCP ; Y TCP ; Z TCP ; CTCP ; qTCP ; wTCP T _

Depending on geometric parameters, such as number of joints, length of arms and distance between supporting points, the analysis rst leads to the workspace of the structure. By calculating from estimated backlash of ball bearings and resolution of the robot drives systems, the matlab-analysis shows the forecasted sensitivity of the structure. Sensitivity E is dened as the ratio of change in the output signal to the change in the input signal: DX O DX I

DX O output; DX I input

In general, sensitivity can be subdivided into sensitivity of structure _ and sensitivity of dimension tolerance J . J _
ME

68

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Table I Requirements
Criterion Workspace Footprint (area of robot base) Theoretic resolution Linear speed Value , chipcard-size (80 50 mm) , DIN-A4 (210 297 mm) Up to 1 mm or better . 100 mm/s

2 J _

r _

6 6 . 6 . . q 6 4 w _

x q1

.. .

x q6

3 J _
ME

2 x

q1

7 . 7 . 7 . 7 5 w

r _

6 6 . 6 . . _ 6 l 4 w

l 1

.. .

q6

l 1

7 . 7 . 7 . 7 5 w
l i

x l i

with a different ratio of arm 1 compared to arm 2. The serial structure has a sensitivity map that only achieves good values in very limited areas of the available workspace. The parallel structure, in comparison, achieves good values over almost the complete workspace and also has a symmetrical sensitivity map. Furthermore, the parallel structure should offer a signicantly better dynamic performance, because only the gear motor for the fourth (rotational) axis is carried by the moving arm. Additionally the passive joints of the parallel structure are easier to miniaturise than active joints. Because of these reasons, the parallel SCARA structure was chosen as the basic structure for further developments. An argument against the parallel structure would be its more complex control due to the closed kinematic structure. Nevertheless, the steady decrease in the cost of computing power means that in this case the mechanical advantages of the parallel structures are more important.

Being q the robot drives and l its arm lengths. As can be seen in the elements of the matrix J, sensitivity of structure _ _ provides information about the deection of the robots pose caused by false positions of the drives q. The elements of the matrix J , depending on the robots arm lengths l, provide _ ME information about the pose deections caused by elongation of robots arms, resulting from temperature effects, backlash and inaccuracy due to the fabrication and assembly of the robot. Contrary to the sensitivity, the repeatability of a robot structure can only be estimated but not calculated. Based on several estimated interference factors, the results of the sensitivity of the structure and the sensitivity of dimension tolerance can be calculated and serve as an estimation for the theoretical repeatability of the robot. The iterative analysis of different geometrical parameters aimed at a minimised arm length by keeping a chip-card-sized workspace and high accuracy. The theoretical repeatability, based on accuracy data of drive components and ball bearings, is in some areas better than 1 mm. The sensitivity plots of the optimised geometrical structures for serial and parallel kinematics, meeting our requirements, respectively, are shown in Figure 5. Figure 5(a) shows the upper half of the workspace and the sensitivity plot of a miniaturised serial SCARA robot. Figure 5(b) shows the workspace and the sensitivity plot of a miniaturised parallel SCARA with similar arm length as the serial SCARA. Compared to the commercial robot Melfa by Mitsubishi, the arm lengths of the miniaturised structures are scaled down by approximately the factor 4 but Figure 5 Positioning sensitivity of miniaturised structures

6. Description of nal robot design

The results of the design process concerning the basic structure lead to a typical parallel structure shown in Figure 6. Both active joints A1 and A2 are equipped with Micro Harmonic Drivew gears combined with Maxon electrical motors (q1, q2). They drive the plane structure in x-ydirection, whereas joints B1, B2 and C are passive in this case. The plane parallel structure offers two translational DOF in the x-y-plane. The z-axis is integrated as a serial axis in the base frame of the robot. The easy handling of the whole plane parallel structure driven in z-direction is possible because of its minimised drive components and light aluminium alloy structure. The z-axis is driven by a Harmonic Drive servomotor combined with a conventional ball screw. Additionally the rotational hand axis C was designed as a hollow rotational axis integrated in the passive joint C as the tool center point (TCP) of the parallel structure. This allows media like, e.g. vacuum to be passed along the hand axis. A Micro Harmonic Drivew gear combined with a Maxon brushless motor drives the C-axis. This axis with a diameter of 2.5 mm can be equipped with a vacuum gripper. All joints of the robot consist of preloaded angular contact ball bearings that are strained by springs. The resulting joints are nearly free of backlash and have low friction. All servomotors support a high resolution encoder feedback signal. The drives (q1, q2) of the parallel structure are equipped with encoder resolution of 256 increments with a gear ratio of 500:1, so that the angular resolution of the robot

69

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Figure 6 Parallel structure of the robot

A matlab-analysis based on the kinematic transfer functions mentioned in section Robot design approaches led to a linear speed diagram shown in Figure 10.

7. Measurements
Performance specications regarding accuracy were measured at the IWF. The applied test method conforms to the ISO standard EN ISO 9283. This means that the repeatability was measured 30 times at ve denite points within the rectangular workspace shown in Figure 10. The results of the repeatability of this rst prototype are shown in Table III. Figure 11 shows the plots of the best and the worst results. As expected, the practically measured repeatability is lower than the forecasted high theoretical accuracy. Experiments demonstrate that elasticity in the drive components may be responsible for this effect. Friction in the ball bearings of the structure can lead to a deection of the micro gear. The encoder at the input of the drive cannot detect this deection on the output of the gear, so this deection cannot be compensated by the robot control.

arm is of 0.0028 degrees. Additionally the Parvus is equipped with magneto resistive position sensors for the initialization of the robot as well as an emergency stop function. The control of the Parvus robot was developed on a realtime system from dSPACE, Inc. (Germany). The system features a PowerPC750 digital signal processor (DSP) running at 480 MHz, a digital I/O board, high resolution analog I/O boards, an encoder board and a serial I/O board. The dSPACE system was chosen because it is a powerful hardware solution and operates with the communication software Control Desk which was used to create the graphical user interface. For programming the control, dSPACE, Inc. provides the possibility to use program codes in Matlab/Simulink and C. Figure 7 shows the basic experimental setup for the robot and the control. The rst two functional prototypes of Parvus, shown in Figure 9, are currently being used for further research and presentation purposes. As demonstration application, the robot actually picks and places glass balls with a diameter of about 1 mm (Figure 8 and 9). The rst prototype of Parvus fulls the following specications shown in Table II. Figure 7 Experimental setup

8. Conclusion and outlook

The objective of the Parvus was to show the possible degree of miniaturisation of a robot structure with a robot design similar to conventional robots. The Parvus is equipped with all the functionalities of a pick-and-place-robot and lls the gap between commercial precision robots and designs for millimetre-sized micro robots. The two basic robot structures of a miniaturised serial SCARA and parallel SCARA were analysed with regard to the workspace and sensitivity data. The parallel structure was chosen because of better results of the estimated repeatability based on the simulation data. The aim was to design a miniaturised, compact, high precision robot with good access to the workspace.

70

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Figure 8 Pick-and-place application of glass balls ( 1 mm) with Parvus

Figure 9 The rst two functional models of the Parvus

Table II Technical specications of the rst prototype

Criterion Workspace (xy, absolute) Workspace (max. cubical) Footprint Robot cell Resolution max. (xy-plane) Repeatability (best, worst) Linear speed Rotational speed (C-axis) Angular resolution (C-axis) Payload Notes: aRatio 160:1; bratio 500:1 Value 4658 60 45 20 100 53 130 170 ,1 5.9, 14.1 .100 187a/60b 0.022a/0.007b 50 Unit mm2 mm3 mm2 mm2 mm mm Mm/s rpm 8 g

First measurements of the robot conrmed the approach. The rst functional model is several times smaller than conventional precision robots such, e.g. the Melfa by Mitsubishi. The measured repeatability is in the same range as that of conventional systems, but provides further potential for being optimised. Concerning the accuracy parameters of the robot, there will be further efforts to optimise the robots structure and drive mechanism. Meanwhile, the micro-gears are available with an optimised tooth prole, better accuracy and higher stiffness. Hence, the improved Parvus will guarantee a much better repeatability than 10 mm. The prototype of the robot (Figure 8) was demonstrated to the public for the rst time at the Hanover Fair in April 2005. The good feedback from industrial partners provides prospects for using the Parvus in industrial applications or a visionary desktop-factory. 71

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

Figure 10 Linear speed diagram and measured workspace of Parvus

Table III Measurements of repeatability in xy-plane

Point P1 P2 P3 P4 P5 Repeatability 9.1 mm 10.1 mm 5.9 mm 7.1 mm 14.1 mm Standard deviation 2.1 mm 2.5 mm 1.3 mm 1.5 mm 3.3 mm

Figure 11 Results of repeatability of the rst functional model

72

PARVUS miniaturised robot Arne Burisch et al.

Assembly Automation Volume 27 Number 1 2007 65 73

References
Breguet, J.-M. and Bergander, A. (2001), Toward the personal factory?, Proc. of SPIE, Microrobotics and Microassembly III, Vol. 4568, pp. 293-303. Gaugel, T., Bengel, M. and Malthan, D. (2003), Building a mini-assembly system from a technology construction kit, Proc. of International Precision Assembly Seminar IPAS2003, Bad Hofgastein, Austria. Hesselbach, J., Pokar, G., Wrege, J. and Heuer, K. (2004a), Some aspects on the assembly of active micro systems, Production Engineering, Vol. XI/1, pp. 159-64. Hesselbach, J., Raatz, A., Wrege, J. and Soetebier, S. (2004b), Design and analysis of a macro parallel robot with exure hinges for micro assembly tasks, Proc. of 35th International Symposium on Robotics (ISR) No. TU14-041fp, Paris, France. Hesselbach, J., Wrege, J., Raatz, A. and Becker, O. (2004c), Aspects on design of high precision parallel robots, Journal of Assembly Automation, Vol. 24 No. 1, pp. 49-57. Koelemeijer, S. and Jacot, J. (1999), Cost efcient assembly of microsystems, MST-news, No. 1, pp. 30-2.

Okazaki, Y., Mishima, N. and Ashida, K. (2002), Microfactory and micro machine tools, Proc. of KoreanJapan Conference on Positioning Technology, Daejeon, Korea. Rochdi, K., Haddab, Y., Dembele, S. and Chaillet, N. (2003), A microassembly workcell, Proc. of International Precision Assembly Seminar IPAS 2003, Bad Hofgastein, Austria. Tanaka, M. (2001), Development of desktop machining microfactory, RIKEN Review, No. 34.

Further reading
Burisch, A., Wrege, J., Soetebier, S., Raatz, A., Hesselbach, J. and Slatter, R. (2006), Parvus a micro-parallel-SCARA robot for desktop assembly lines, Proc. of IPAS 2006, 19-21 February, ISBN-13: 9780-387-31276-5, Bad Hofgastein, Austria, pp. 65-74.

Corresponding author
Arne Burisch can be contacted at: a.burisch@tu-bs.de

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com Or visit our web site for further details: www.emeraldinsight.com/reprints

73