Micro Robot for Rotary Desktop Assembly Line

Philipp Kobel and Reymond Clavel Institute of Microengineering - Laboratoire de Système Robotiques, STI - LSRO

École Polytechnique Fédérale de Lausanne, EPFL Lausanne, Switzerland

Philipp.Kobel@epfl.ch and Reymond.Clavel@epfl.ch

Flexible and miniaturized production lines, known as desktop-, pocket- or micro-factories, with an integrated cleanroom are needed to meet the performance requirement of today’s microproducts. In respect thereof, a miniaturized, large workspace robot is developed for microassembly tasks. Improvements based on recent benchmark tests and simulations are rounding out the overall paper.

Micro-factory; cleanroom; micro assembly robot; tool changer; finite element analysis

I. INTRODUCTION Since the introduction of the first microsystems a few

decades ago, significant progress has been achieved in the fabrication of microdevices, but the assembly and packaging at the mesoscale and microscale still impose a major obstacle to their commercialization. For this reason, on an industrial scale, these backend processes account for 60% to 90% of the overall production costs [1].

Although microscale robotic assembly, arbitrarily defined as manipulation of sub-millimeter components, with accuracies typically ranging from 0.1 µm to 10 µm, shares many common aspects with the traditional robotic assembly systems. However, there are some particularities of the microworld that have to be considered when handling microcomponents. Below one millimeter, the law of gravity becomes insignificant compared with surface forces, since electrostatic, capillarity and Van der Waals’ forces start to predominate at such scale [2]. Moreover, the manipulation of objects with micrometric structures requires special caution, because these high technology products are often fragile and vulnerable to scratching, deformation and contamination caused by microparticles. As a result, many microtechnology products must be handled in a very repeatable temperature-, humidity- and microparticles controlled environment. Accordingly, they typically require large cleanrooms and thus are normally associated with enormous construction and operating costs. Furthermore, conventional industrial equipments are rarely optimized to efficiently meet the accuracy requirements of the microworld. For this reason, the downscaling of production facilities is required and has extensive implications for the manufacturing concepts.

This contribution addresses these issues and aims to integrate microrobots into miniaturized manufacturing facilities in order to meet the performance for miniaturization. As the miniaturization goes hand-in-hand with the necessity of a clean and well-controlled environment, the compatibility with the

cleanroom and the integration of the environment control into the micro-factory (mini-environment) are crucial points. Especially, as they reduce the cleanroom to the absolute minimum and allow its operation outside of expensive cleanrooms. These far-reaching modifications of the traditional production concept have triggered the authors’ effort to develop an appropriate manipulator. Because this particular working environment is the main driving force of this robot, this paper starts by describing such a micro-factory, before it presents the technical refinements and properties of this robot.

II. MICROFACTRORY CONCEPT

A. System overview In former works, the authors’ laboratory LSRO has

developed two micro-factory concepts: A linear assembly line (in collaboration with the Centre Suisse d’Electronique et de Microtechnique S.A. CSEM) [3] and a circular production chain [4]. The linear approach is composed of modular units called microboxes (Fig. 1), whereas in the circular approach the production cells are located around a central unit. The diameter of this circular micro-factory is about 1 m. Both concepts combine the performance of a miniaturized environmental control and desktop manufacturing units, which are summarized by the following three key elements:

• Size-reduction: An advantageous approach to improve the accuracy is to reduce the overall size of the assembly line, namely smaller control-, energy- and force loops. Furthermore, the size reduction of the assembly line optimizes the disproportional size ratio between microproduct and production equipments and improves the motion dynamic due to smaller inertia and distance of travel, as well as thermal behavior.

• Mini-environment: The small, enclosed workspace with clean air environment is very economical, as it reduces the cleanroom to the absolute minimum, is faster to configure (high air renewal guarantees decontamination delays less than a couple of minutes) and it works stand-alone in normal ambient air.

• Modularity: A major focus on the design of a modular construction kit permits the rapid customization of the production chain in order to meet the requirements of a broad diversity of assembly task. The reconfiguration of microscale assembly cells is even easier to handle and therefore reduces the necessary setup time.

Although the linear and circular micro-factory concepts share these advantages, there is a significant difference between these two approaches: The circular assembly line (Fig. 2) eliminates some discontinuities of the clean air flow that occurred in the linear setup. This improvement allows a more uniform air renewal and significantly reduces the risk of particle contamination. In addition, it offers an optimized conveyor without affecting the modularity of the system. For this reason, the following discussion of this contribution focuses exclusively on the circular disposal.

B. The production cell in the circular micro-factory In the novel concept, the production modules are circularly

arranged around a central unit forming a production chain. This central unit harbors the common features, such as the global rotary transfer system and the circular air inlet. The conceptual layout thus allows uncoupling the common features in order to get well-separated production cells, assuring “plug-and-produce” regardless of the central unit and neighboring cells. The benefits of a cellular modularity are obvious: faster reconfiguration, the possibility of remote calibration and a library of interchangeable modules.

Apart from the aforementioned flexibility of the production chain, a special attention is given to provide a submodular customization of the production cells. On the one hand, the

manipulations are done on 2-inch trays improving the industrial cooperation by standardizing the production flow. On the other side, there are predefined areas for the process tools in order to guarantee an optimal utilization of the available module envelope. Each production module can be individually equipped with one or several features, e.g. the module’s ground plate can harbor feeding-, tool holder- and referencing systems, input/output SAS for 2-inch trays (Safety Access System (airlock system) allows linking the clean inside with the boxes, which are used to transport the product in the ambient outside). The large workspace manipulator will be incorporated in the ceiling, as this emplacement offers the best conditions for accessing the rotary table, the sidewalls and the ground plate of the production cell. The design of the robot is strongly influenced by its working environment in the production cell, which summarized as follows:

• Work space with clean air environment (class 1 to 10000 or 3 to 7 after ISO 14644-1) due to the central inlet generating a 360° laminar clean airflow without discontinuities. The zone between the rotary table and the air outlet behaves in a similar way as laminar flow cabinets allowing a complete isolation of dirty operations from the neighboring cells.

• A climate control (temperature, humidity, pressure, special gas, air sterilization). The temperature stability is of particular importance to reduce the thermal drift during high-precision manipulations.

• Access to the global rotary transfer system and compatibility with the production under clean conditions.

• Standardized 2-inch carrier providing a high compatibility with industry standards that can easily be handled by the manipulators, the conveyor or the SAS.

Despite the seemingly straightforward development of a micro-factory manipulator, its miniaturization and the radial clean airflow require consideration of additional criteria, beyond those associated with the former prototype and traditional robotics. The main challenges are the lack of standardized micro assembly robots, the compatibility with the circular cleanroom and the full exploitation of the workspace. Therefore the LSRO develops a manipulator adapted to the micro-factory’s potential for a broad diversity of assembly task in a flexible micro-factory production chain.

III. ROBOT DESIGN

A. Conceptual constraints For the basic manipulation in the micro-factory, a robot

with at least four degrees of freedom (x, y, z, Φz) and with a tool changer at the end effector is needed. In order to be compatible with the aforementioned concept, the manipulator has to fulfill the following key restraints:

• Cleanroom compatibility: Passive (material, form etc.) and active (e.g. exhausting polluted air) preventive measures to reduce the impact of the handling system on the cleanroom.

Figure 1. CAD model with linear arranged production modules called microboxes (left) and real prototype (right).

Central unit 1 Clean air inlet (Van, HEPA

filter, humidity and temperature control)

2 Rotary table 3 Standardized 2-inch trays 4 Interface for the modules

(mechanic, electric and of data)

Surrounding production modules5 Space for manipulator 6 Safety Access System SAS

(Inlet / Outlet of components) 7 Working volume with laminar

and horizontal air flow 8 Tray-referencing system 9 Air outlet

Figure 2. Overview over the concept of the rotary assembly line with four modular production cells of the usual six.

9

7 65

4

83

1

2

• Flat robot design: The low-ceilinged shape of the production cell is needed to compensate for the radial expansion of the clean airflow. Therefore, the robot thickness has to be adapted, as its height reduces the residual working volume and represents a flow obstruction.

• Ratio “footprint-workspace”: The manipulator has to cover the whole working volume of the production cell in opposition to the base of the robot. Its footprint should be as small as possible, in order to free space for the visual system, tool magazine, feeders, etc.

• Modularity: The robot has to be easily exchangeable, due to the flexibility requirement explained in the previous chapter. Moreover, the robot has to adapt itself to a certain number of assembly tasks and therefore requires a tool changer at the end effector.

B. The robot “ΦR” and its kinematic structure A theoretical analysis by the LSRO revealed that a two-

armed robot combines the best performance of a flat design, rigidity and slim form for aerodynamics. These two arms are mounted on a rotary stage Φ and are only used for a pure radial motion R (hence, the name “ΦR”, which is pronounced like the number “4” in German, corresponding to its degrees of freedom). This configuration uncouples the rotary and radial motions and is the main difference compared to the similar robots Parvus or Apis [5] and the H-SCARA of Tampere University of Technology [6]. In fact, the kinematics of the ΦR robot can be divided into three stages:

• The first stage is integrated in a cylindrical structure, allowing the vertical translation in z and the rotation around its main axis Φz (Q1 and respectively Q2 in Fig. 3). Only the bottom of the cylinder extends into the cleanroom.

• The second stage is the double-arm structure mounted under the cylinder with an offset from the center. The two upper arms of the parallel structure are inversely linked (indicated by Q3 and –Q3), which leads consequently to a pure radial movement. Thanks to the offset, the end effector (namely the third stage) can reach the center and forms a complete cylindrical working volume.

• Finally the end effector is the arbor connecting the two arms together. The actuating mechanism for the associated rotation Q4 is integrated in the thicker hollow arm.

By the virtue of the joint orientation in the double-arms, this type of structure benefits from the same remarkable advantages of the classical Selective Compliance Assembly Robot Arm (SCARA) structure. The controller for the X-Y direction can be configured more compliantly than the Z direction, hence the term “selective compliant”. This is advantageous for many types of assembly operations, e.g., self-alignment due to a selectively compliant robot during the insertion of the pin into a hole.

To conclude, this double-arm structure is part of a coherent strategy for a cleanroom compatible design [7]. An appropriate additional engineering effort allows placing most of the critical elements, such as motors, gears, bearings, controllers etc., outside of the cleanroom. This approach allows bypassing, to a certain degree, the lack of microscale technology and the cleanliness constraints. Therefore, off-the-shelf components can be used by exteriorly concealing them in the housing body of the production cell. Parallel structures are especially well adapted, as their actuators can be placed on the base and only the passive structure interacts with the clean environment. Furthermore, the two hollow arms contain the power and data cables and the conduits for the vacuum or compressed air supply for the tool holder. The mechanism of the end effector is embedded in the cavity of the thicker arm. A negative pressure gradient between the clean environment and the hollow arms inhibits the leakage of microparticles. In a similar manner pre-stressed ball bearings are embedded behind non-contacting gap seals in robot structure.

Q1Q2

Q3-Q3

Q4

Figure 3. Kinematic structure of Robot ΦR.

1 First cylindrical stage (Q1 and Q2) 2 Upper arm 3 Forearm with integrated motor

and sensor for the Q4 4 End effector and tool changer

5 Ceiling for radial airflow compensation

6 Ground plate 7 Workspace in the production cell 8 Range of manipulator in the open

space

Figure 4. A single production cell with robot ΦR and its cylindrical working range and working volume in production cell (almost rectangular volume).

76

5

483

1

2

C. Robot work range and workspace in the production cell To facilitate the manufacturing, the generic concept houses

six similar fabrication modules of about 250 mm x 250 mm footprint and 75 mm height. One production cell with an integrated robot ΦR is shown in Fig. 4. The working volume in the production cell is about 50 mm smaller in x and y and only 45 mm high due to the space requirement of the robot, as symbolized by the almost rectangular volume. The working volume is only a subset of the complete working range (cylindrical shape) of the manipulator.

D. Tool changer A combination of a suction cup and integrated positioning

system emerged from the evaluation of several design approaches for a tool changer. In this solution (Fig. 5), an ordinary o-ring seals the air gap and three positioning balls fit into the v-grooves of the counterpart. The best performances (accuracy and tractive force) are achieved, when they are located as externally as possible. Furthermore, this solution leaves the central space of the tool holder unused, so larger tools can be implemented between the few millimeters thin tool holder.

A prototype of the tool changer (Fig. 5) has been fabricated in the author’s laboratory in order to quantify precisely the technical specifications. The benchmark setup includes an end effector that was rigidly fixed and connected to the vacuum pump and a second part to represent the tool holder. In the experimental setup an ordinary o-ring with a hardness of 60 Shore A seals the gap and can be later easily replaced by cleanroom compatible counterpart. The mounted o-ring is floating and a simple bulge on the end effector (in the test setup three overlapping wedge-shaped pins in rectangular grooves) prevents the o-ring from falling off. This setup efficiently allowed the measurement of the maximal axial holding force Fa, mes as a function of the applied depression using a dynamometer.

The results are shown in Fig. 6A. The holding efficiency is the percentage rate of the measured force to the calculated axial holding force Fa, cal (based on the surface and the depression). It can be seen that the o-ring does not tighten at weak depression. The holding efficiency of the tool changer with a surface finish of N6 (maximal roughness Ra of 0.8 µm) quickly reaches the

values given by the manufacturer of 60% to 80%, resulting in an admissible lifting capacity up to 10 kg.

Further calculations based on the measured axial holding force (Fig. 6B) reveal additional proprieties of the tool changer. The radial carrying capacity Fr, cal is about three quarter of the lifting capacity. The radial carrying capacity and twisting moment Ma, cal are remarkably high thanks to the interlocking of the positioning balls with the v-grooves. The tilting moment Mr, cal is also optimized due to the maximization of the suction capacity and spatial separation of the balls, which enhances the positioning accuracy even more. It is needless to say that the three balls in the v-grooves form one of the most accurate and reproducible positioning methods, as they establish exactly the six punctual references at the minimum energy position in the v-grooves to block all degrees freedom.

In conclusion, the proposed layout satisfies in most aspects: Its flat design allows larger tools and its carrying capacities overpass the requirements of micro-assembly tasks. Moreover, it guarantees a fast and reliable behavior, maximized suction and repeatable positioning. Last, but not least, the lateral overlapping of the o-ring offers some interesting additional advantages. First, as a locally unsealing of the o-ring substantially reduces the holding force. Hence, it simplifies the manual removal of the tool. Second, it could also work in the same way as an anti-shock device. A lateral collision unblocks the tool holder and the inertia of the manipulator does not cause a cumulative impact on the tool holder with its fragile microtool.

1 Connection to the vacuum pump 2 Tool holder 3 End effector 4 Floating mounted o-ring

5 Three 120°-arranged balls6 V-groove 7 Rectangular groove for o-ring

holder Figure 5. Schema of the tool changer (left) and benchmark setup for

characterizing the tool changer (right).

0

20

40

60

80

100

0

20

40

60

80

100

120

140

-1000 -900 -800 -700 -600 -500 -400 -300 -200 -100 0

Hol

ding

eff

icie

ncy

[%]

Axi

al h

oldi

ng f

orce

[N]

Depression [mbar]

Fa, mes [N]

Fa, cal [N]

Holding efficiency [%]

0.0

0.5

1.0

1.5

2.0

2.5

0

20

40

60

80

100

120

140

-1000 -900 -800 -700 -600 -500 -400 -300 -200 -100 0

Adm

issib

le to

rque

[Nm

]

Adm

issib

le f

orce

[N]

Depression [mbar]

Fa, mes [N]Fr, cal [N]Ma, cal [Nm]Mr, cal [Nm]

Figure 6. Characteristic of the tool changer: (A) axial holding force and efficiency and (B) admissible forces and torques.

45mm

Fa, mes [N]

Fa, cal [N]

Holding efficiency [%]

Fa, mes [N]

Fr, cal [N]

Ma, cal [Nm]

Mr, cal [Nm]

B

A

7

65

4

31

22

1

4

537

6

IV. FINITE ELEMENT SIMULATION

A. Modelization of robot structure The Finite Element Analysis (FEA) was conducted using

PTC's Creo Elements/Pro Mechanica. The simulation results were used to test and optimize the structural performance and the dynamical proprieties of the concept.

The computer-aided design (CAD) data of a slightly simplified robot structure were employed to get a three-dimensional FEA model with the most realistic behavior. In the same way, each articulation of Fig. 3 with a pair of preloaded and o-mounted bearings was substituted with idealized bearing models in order to account for the influence of the bearings. Fig. 7 represents an idealized bearing, where spring connections between inner and outer ring form the equivalent stiffness of the real bearings. The radial and z-component of the spring constant were calculated using (1) and (2), where Nsprings must correspond to at least a number of two orthogonal springs. The simulations were conducted using an axial and radial rigidity kaxial and kradial of 3.3 N/µm and 20 N/µm, respectively. These values correspond to bearings with an interior diameter of 25 mm, 4 mm height, twenty-one 1/16-inch balls and a contact angle of 15°.

Special emphasis was made to disperse the punctual stress concentrations due to the point-to-point spring connections. Accordingly, six springs, more than the essential number of springs, were implemented and, in addition to this, a nonexistent high elastic modulus was assigned to the two rings. This approach was feasible, because the ring rigidities had already taken into account by the spring connections.

(1)

(2)

The idealized passive articulations have no angular rigidity (consequently the tangential spring component kφ is also zero), whereas the actuated joints (±Q3 and Q4 in Fig. 3) have an angular rigidity given by the gearing and the controller. However, in the simplified model only the angular rigidity of the belt-driven pulleys was taken into account. Due to the pre-stress of the belt, the angular rigidity of the pulley depends on the belt stiffness at both sides kbelt1 and kbelt2 (tight and slack span) as shown in (3).

(3)

For this reason, with respect to the belt, the same angular rigidity is applied on the arbor.

B. Static behavior To simulate the static behavior, an external load was

imposed at the tool holder inducing an elastic deformation. The rigidity of the system was evaluated by the ratio of the applied force and the simulated displacement. The results are given in Tab. I and illustrated with an amplified deformation.

It can be seen that the behavior is only slightly asymmetric. In fact, forces in the plane of the double-arms cause essentially traction and compression in the forearms and therefore the profile of the thinner arm is quite well adapted and does not provoke any serious drawbacks. Furthermore, off-plane forces primarily afflict the main arm. The static analysis in z reveals that a weak point is situated at the elbow articulation. Additional effort has to be invested to find the best compromise between stiffness and spatial requirement. This is especially challenging, as the overall height reduces the short working range in z and as cables for power, data and vacuum have to pass through the joint of the double-arms.

In conclusion, the results are promising, as the stiffness is adequate for the weight of the transported microdevices or for the forces of the micro assembly operations and it is also in proportion to the resolution of the selected sensors. Moreover, the thinner arm is an interesting method to avoid a hyperstatic system, to get a collision-free motion with the double-arms and to gain more space for the tool changer.

Figure 7. Idealized bearing model with 6 springs.

TABLE I. STATIC ANALYSIS OF ROBOT ΦR

Direction of applied

force

Stiffness of robot ΦR

Illustration Rigidity [N/µm]

r 0.41

φ 0.14

z 0.10

kaxial

kradial kangular

kr kz

kφ

C. Dynamic behavior The same model allows computing the eigenfrequencies of

the robot. The initial simulation revealed two eigenfrequencies under 100 Hz. The aforementioned modes were both in the plane of the double-arm, namely in R- and φ -direction, caused by the low angular rigidity of the upper arms. The first eigenfrequency that was independent of the rigidity of the upper arms, was 2.3 times higher than the lowest one. Therefore, an effect was invested to boost the lowest eigenfrequency. A factor of 5.3, according to the square, is needed, as the eigenfrequency varies proportionally to

(4)

where k is the rigidity and M is the mass. The first simulation was carried out with a radial stiffness of the upper arms corresponding to a Synchroflex® timing belt with a standard profile T2.5 and belt width of 10 mm. The profile T5 and belt width of 16 mm fulfills almost the criterion, which was successfully confirmed by the control simulation. Both results are represented in Tab. II. An interesting and contradictory point is that the smallest timing belt, respecting tensile load, rotation speed, number of meshing teeth and so on, may allow the most effective dynamic, but this case shows that it is even better to oversize the belt in order to meet the better vibration characteristics. Consequently the FEA delivers an excellent harmonic behavior needed for ambitious assembly operations.

V. CONCLUSION The cinematic structure and the first results of the novel

assembly robot “ΦR” are presented. The cylindrical manipulator was developed to perform a wide range of assembly tasks in the context of a flexible micro-factory production chain with an integrated, clean and controlled environment, which has a great technical and economical potential compared to the current bulky assembly lines placed in cleanrooms of ballroom size. The robot design has therefore to fit the strictest cleanroom constraints, size limitation, as well functionality and modularity requirements.

The reduction of size of the robot has proven to be an elegant line of attack to increase accuracy, flexibility, and dynamics, as well as to make the manipulator compatible with the constraints of a mini-environment. As the housing body of micro-factories remains relatively large in comparison to the miniaturized cleanroom, the main strategy to reduce the impact of the handling system on the cleanroom was to hide most of the manipulator’s structure in the housing body. This external location not only rendered the requirements for size and cleanliness of the critical mechanical parts in micro-handling systems less rigorous but also helped to bypass the lack of technology at microscale, which currently limits the emergence of automated micro-handling machinery.

Beyond the environmental constraints, the structure of the robot allows a relatively large workspace for a small footprint and respecting the low ceiling of the production cell. In order to guarantee a high flexibility of the manipulator, a special

emphasis was given to the development of a tool changer. The result combines a flat functional design, high performance and accuracy. Static and modular analyses of the manipulator revealed promising results and means to further improve the design. Before fabricating a prototype, some more research is needed, for example, a benchmark test to find the ideal air gaps and an optimization of the pressure gradient inhibiting. Together with the integrated cleanroom, these measures will eventually lead to a fully functional micro-factory.

ACKNOWLEDGMENT The authors would like to thank Dr Irène Verettas and

Stephane Rossopoulos for the conceptual framework and valuable ideas. Particular thanks to Bandar Hakim and Stefan Kobel for proof-reading this manuscript.

REFERENCES [1] D. O. Popa, and H. E. Stephanou, “Micro and Mesoscale Robotic

Assembly,” Journal of Manufacturing Processes, 2004, Vol. 6, No. 1, pp. 52–71.

[2] M. Dafflon et al, “Characterization of Micro Manipulation Tasks Operated with Various Controlled Conditions by Micro Tweezers,” Preceedings of the 5th International Workshop of Micro-factories, 2006.

[3] I. Verettas, R. Clavel and A. Codourey, “Micro-factory: Desktop Cleanrooms for the Production of Microsystems,” Proceedings of the 5th IEEE, International Symposium on Assembly and Task Planning, 2003, pp. 18–23.

[4] P. Kobel and R. Clavel, “Circular concept of a miniaturized assembly line with an integrated cleanroom,” Preceedings of the 7th International Workshop of Micro-factories, 2010, pp 25-29.

[5] A. Burisch and A. Raatz, “Challenges of Miniaturized Robots and Machine Elements for Desktop Factory Applications,” Preceedings of the 7th International Workshop of Micro-factories, 2010, pp 100-105.

[6] A. Vuola et al, “Micro H-Scara Robot: Findings and Results,” Preceedings of the 7th International Workshop of Micro-factories, 2010, pp 20-24.

[7] P. Kobel and R. Clavel, “Miniaturization Challenges and Their Impact on the Micro-factory Concept and Manipulators,” Journal of the Japan Society for Precision Engineering, 2011, Vol. 77, No. 3, pp. 263–268.

TABLE II. HARMONIC ANALYSIS OF ROBOT ΦR

ModeNo.

Eigenfrequencies of robot ΦR Timing belt T2.5, 10mm width Timing belt T5, 16mm width

Description Freq. [Hz] Description Freq. [Hz]

1 66 145

2 98 153

3 152 220

4 550 557