1/Assembly

Introduction
Today, approximately 70 years after the assembly line invention, the interest in automated assembly is still growing because of the high manual labor content of most complex assembly operations, and the need of flexible assembly systems as products become more complex and various.
The term assembly means the fitting together of two or more discrete parts to form a new subassembly. Handling and orienting parts are the main tasks in assembly, but it also includes mechanical fastening operations using screws, nuts, bolts, rivets, etc.
Robot assembly offers an alternative with some of the flexibility advantage of people and uniform performance of fixed automation, but not without failures and weakness sides. The economic issues in automated assembly applications concern generally management's objectives for the product, unit cost and allowed investment, while technical issues deal with performance, traditional part-mating or assembly sequence problems.
An assembly cell is composed mainly by robots, sensors, vision system, feeders and conveyors (Fig. 1). The performance of the whole cell depends on the performance and the suitable choice of each of its equipment. For assembly robots, performance is highly tied with accuracy, compliance, velocity and repeatability characteristics. In addition to performance, an assembly cell has to be flexible by offering the possibility to assemble an arbitrary combination of products and to adapt to frequent changes in product design.
All these elements are actually parts of many research projects,  in the goal to reach and exceed the human ability and dexterity [Nof85] & [Gro86].

 
Figure 1: The MARK IIF Flexible Automatic Assembly Cell assembling 60 variants of air-motors at Atlas Copco
                    [And96] .

1.1/Peg-in-hole issues [Gro86]
The round peg-in-hole application needs only 5 degrees of freedom, while the square peg-in-hole mating needs 6 in order to adjust the corners of the square peg with the corners of the hole (Fig. 2), this case concern especially electronics industries, deburring and stacking assembly operations. The majority of the robotic automation tasks, require constrained motion with both position and angular control, which mean that compliance only in the lateral directions is not enough.

Figure 2: Round and square peg-in-hole mating [Gro86].
1.2/Passive accommodation [Nof85]
A full directions compliance can be accomplished by a passive accommodation in which reaction forces and torques as sensed by a compliant wrist are used for the correction of residual positioning errors.
The Remote Center Compliance (RCC) device is often used for misalignment compensation during automated assembly.
The Compensator allows an assembly machine to compensate for positioning errors due to machine inaccuarcy, vibration or tolerance, thereby lowering contact forces and avoiding part and tool damage. When the remote compliance center is near the contact point, the part will align with the hole automatically; correcting lateral and rotational misalignment (Fig. 3). The Compensator is a mechanical device that uses high-quality elastomer shear pads to control the compliance.
Figure 3:  RCC device ATI Industrial Automation,
 Internet (http://www.thomasregister.com/olc/ati/another.htm).
1.3/Active accommodation [Ang97]
Compliance can also be accomplished by an active accommodation considering a relationship between the interaction forces and the position. By controlling the manipulator position and specifying its relationship to the interaction forces, it is possible to ensure that the manipulator complies with the constraints imposed by the environment while maintaining appropriate contact forces.
A program is necessary to drive the robot with trajectory and force motion inter-relationships inputs. This application uses a 6 axis force-torque sensor  to measure the interaction forces. The necessary compliance is achieved by modifying the motion of the manipulator according to a programmed force-deflection relationship at the tool tip.
The objective of the proposed active compliance control scheme is to make the manipulator exhibit a behavior similar to that of the ideal spring: dx = Cf. The required reaction of the robot hand dx is computed from the measured forces. The free space motion trajectory is then modified by adding this offset displacement dx. The desired compliance C is specified according to the task and dictates the desired force-motion inter-relationships. The robot is then commanded to follow the modified trajectory. The algorithm can be easily implemented on current industrial robots without the need to modify the control algorithms already built into the controller. Examples of active compliance control are shown in figure 4.
Figure 4: Peg-in-hole assembly. The robot is able to insert a peg into a hole with small tolerance, typical of an assembly operation. This is typically achieved by a remote centre compliance (RCC) device. But here the active compliance control scheme can emulate an RCC to achieve successful insertion. Internet (http://www.eng.nus.sg/EResnews/May97/may97p10.html).
1.4/Sensors
Sensors are a key part of any robotic system. They are classified by internal state and external state sensors.
Internal state sensors are devices to measure position, velocity and acceleration of the robot joints and/or end effector. These sensors provide feedback for the robot's position and motion control system, as limit switches, potentiometers, optical interrupters and tachometers, and generally they present high resolution.
External state sensors as tactile sensors, force/torque sensors, and vision sensors are mainly used by assembly robots.

1.5/ Force torque sensors [Sch97]
The measurement, validation and output of reaction forces are possible with help of force torque sensors. In connection with robots and their peripheral units, force torque sensors allow a flexible reaction on changing of process parameters. The form changing of the deformation body in fact of the force impact will be detected by means of straingauge to determinate the force (Fig. 5), the complete electronics of the signal processing including a microcontroller is located inside the sensor body (Fig. 6). Therefore force torque sensors open additional applications of robot technology in the field of assembly, grinding, polishing and deburring. In this fields position errors or fabrication tolerances have significant force changes.

 
Figure 5: Force Torque Sensor used on PUMA 500 in peg-in 
              hole application. 
http://www.iff.fhg.de/iff/aut/projects/ kraft/kraft_e.htm		  
 
 

Figure 6: Interior view of a force torque sensor; view of the deformation body.
1.6/Vision sensors
In the case of Assembly, the main application that vision sensors are used for is insertion, as hole detection for peg mating, or screw fastening operations (Fig. 7) [Hor93]. An example of a screw fastening task has the following description of motions:
 
 I)  The camera is moved to over a screw hole of an 
     object. 

II)  The screw hole is scanned and recognised. 

III) When the distance between the centers of upper and 
      lower holes is beyond the criterion, screw fastening is 
      not carried out and a failure signal is sent to the robot. 

IV) When the distance between the center of upper and 
      lower holes is within the criterion, positions correction 
      data is sent to the robot. 

V)  The screw-fastening driver is positioned to the position 
      obtained by adding the correction data to the screw 
      fastening position taught in advance. 

VI)  The driver is lowered to fasten the screw. 

VII) Control returns to I) for the next motion. 

  
Figure 7 : System configuration for visual recognition        equipment for screw fastening robot.
In the case of car industry, generally the measurement of holes employs flying-spot sensors. The flying-spot sensor is based on the tri-angular measurement technique, as shown in Fig. 8. The laser beam is directly at the object to be measured, then the reflected beam is picked up with an optical camera and the position of the object is calculated. The measurement accuracy of the sensor is ±0,1mm [Bad93].
Figure 8: Flying-spot sensor, principle of measurement [Bad93]

Others strategies exist, as for example Self-Calibrating strategy. The key to the strategy is the use of a fixed sensor (cross beam sensor) to localise both a mobile sensor (optical hole sensor) and the peg, while the mobile sensor localises the hole.
The optical hole sensor differs from the cross beam sensor by having both its transmitting and receiving sensor element in the same lens (Fig. 9 & 10). The self-calibrating feature allows to achieve successful dead-reckoning insertions with tolerances of 25 microns without any accurate initial position information for the robot, pegs, or holes. The strategy is fast, the hole localisation can be completed in under a second, holes and pegs are localised in parallel [Pau94].

                                Figure 9: Optical hole sensor                     Figure 10: An instrumented sensor-gripper unit.

Development of vision systems make it possible the success of robots in complex assembly tasks. An example of units to be assembled are water pumps, alternators, and similar in-line assemblies. Robot vision is used to identify parts, either in feeders or coming down conveyors belts, and guide the robot arm in picking up the part; then the part is moved into place, over the product being assembled, and placed in position.
Visual sensors for assembly are often optical transducers such as Vidicon cameras, which translate the brightness of different pixels in a visual scene into a sequence of electrical signals. After an image processing, the image data is transformed into a compact and useful form from which relevant information as position and orientation are extracted.
Many research has been directed towards enhancement of the versatility of machine vision, with a view to minimizing the necessary processing. The Consight system for example, Fig. 11, which one of the first vision integration systems in assembly, uses structured light to cope with visually noisy industrial environment. Two planes of light are focused at inclined angles across a moving conveyor belt into a fixed narrow strip upon which a linear array camera is imaged vertically from above. When an object passes under the beam, the lights are intercepted before they reach the belt surface, causing the beams to be displaced and hence produce a "broken" image in the line camera. By collecting a sequence of these image strips, a conventional two-dimensional image of the passing part can be obtained.

Figure 11: The Consight system

Based on a similar plane-of-light principle, the eye-in-hand camera which is more used nowadays, eliminates the use of a conveyor and is capable of homing in on an object for acquisition, Fig. 12.

Figure 12: Eye-in-hand camera.

Assembly under visual control (http://icsweb.ics.org/ICSInfo/Sample/vision.html)
To guide the assembly manufacturing process, a machine vision can be integrated with
motion control. A vision guided assembly system would include a flexible interface to motion controllers, vision and motion calibrations, vision gauging procedures, and an easy-to-use user interface to set up assembly operations.
The system described below is an integrated vision and motion control system for assembly, based on the AutoVision 90 machine vision system from Acuity Imaging, Inc., Nashua, NH. The system includes a machine vision engine, a motion control interface, and an integrated point-and-click user interface, Fig. 13.

Figure 13: Vision system for assembly.
Machine vision can be used to measure the positions of parts and guide the assembly process to reduce the need for costly and inflexible fixturing. Adding a machine vision system to an automated assembly cell allows it to accomplish all of these manufacturing goals and support such operations as:
· Vision processing to sense part/assembly locations
· Interfacing to motion controllers to move servo/stepper motor mechanisms
· Vision/motion calibrating methods for high accuracy measurements
· Integrating vision and motion user interface to build assembly applications

Vision Processing
Machine vision can be used to locate and inspect parts and assemblies. It can locate parts for pickup, orient parts once they are picked up, find assembly locations and verify assembly operations. A standard, cross-platform application is the ideal tool for controlling vision processing and feedback for motion control.
It is important that the vision software supports many processing algorithms, so the fastest and most accurate algorithm can be used in all cases. For example, printed circuit board fiducials (reference location targets) of various shapes can be located using many different vision processing algorithms, depending on their shape and possible defects. Odd-shaped features can be found (including asymmetric features and ignoring small defects in shape). For example the center of a circle can be located even if part of the circle is missing or intensity varies within and between circles, Fig. 14.

Figure14: Locating the circle center on a circuit board.
As shown in figure 15, a typical vision guided assembly cell contains a moving camera mounted on the mechanism and a fixed location camera aimed upward. The moving camera is used to locate parts for pickup and locate place locations. The fixed camera is used to locate parts once they have been picked up.
Figure 15: Vision mounted system on assembly robot (SCARA).

Parts assembly tasks can be divided into two elements: pickup and placement. Parts are picked up from trays, tube feeders, bowl feeders, etc. Once a part is in the grip of the mechanism's end effector, it is oriented and transported to the place location. With an assembly system such as this, part presentation does not need to be accurate and part orientation may not be important. After a part is picked up, the fixed camera measures its location relative to the mechanism's gripper. The measured offset is called the part frame.

Figure 16: Placement of a part.

As shown in figure 16, placement of a part is accomplished by locating the place location through measurements of local or global features. The offset from the original place location (the location that was originally trained or provided from a database) is called the place frame. Moving the mechanism to a location that includes part frame and place frame will position the part over the place point.

Figure 17: Locating the center and orientation of a part.
The pickup and placement of a chip resistor with a vacuum gripper is a good example of vision guided assembly. Figure 17 illustrates machine vision being used to locate the center and orientation of this part. A vision measurement called a frame (the sideways L with arrows on the ends) is attached to the resistor. The gripper may not pick up the part in the center because of the way it is presented (e.g. in a tray with loose tolerances). The position of the gripper is known from calibration.

When a PC board enters the assembly cell the locations of its fiducials are found with the mechanism mounted camera, defining the place frame. As each component is picked up, it is located over the fixed (up looking) camera and its position and orientation is found. The part and place frames are then used to place the component onto the PC board.
With this procedure, components can be fed from feeders or trays and assembled onto PC boards without the need for high accuracy fixturing. Furthermore, accurate fixturing is insufficient for the placement of large quad pack SMDs (68 lead QFP or larger) or devices with lead pitches less than 50 mils. In these cases, vision guidance is not an option; it is a requirement.
These are the basic system requirements for a vision guided pickup and placement system. Many vision processing algorithms are needed to find part and assembly features. A flexible motion control interface allows a diverse selection of motion control choices. Accurate calibrations are required to locate parts and assemblies. Vision coordinate frames are required to locate, pick up, inspect, find place locations and place parts.

1.7/ Problems related to micro-assembly

The particular problems which occur during the assembly of miniaturised components result from their small dimensions, their high sensitivity to damage and the micron-range precision required in the assembly process. The accuracy attainable by an automated micro-assembly system depends decisively on the ability to compensate for inevitable handling tolerances.

Obstacle facing micro-assembly [Rei97]
In addition to problems related to specific applications, the assembly of miniaturised components exhibits certain special features compared with conventional assembly procedure. The main problems can be summarised in Fig. 18.

Figure 18 :Obstacles to automation in micro-assembly
Tolerance problems
As a consequence of the high precision of miniaturised components, only exceptionally small joining tolerances are permissible. The necessary assembly accuracy, depending on the application, lies between 0.1 and 20 mm. Since the permissible joining tolerances are smaller than the sum of the handling (feeding, gripping and positioning) tolerances, the high assembly accuracy can only be obtained by adopting suitable tolerance equalisation procedures.
The tolerance problems can be overcome by minimising and by compensating for handling tolerances. Measures for minimising handling tolerances are based on shortening the tolerance chain. Examples are the avoidance of gripper changes or the feeding of miniaturised component directly to the effector (e.g. by using a hose), in order to avoid pick-and-place operations. Measures for handling tolerance compensation make use of the forces generated by interactive tactile contact between the components to be joined as a means of generating a compensatory movement (known as passive or compliant system), or else utilise sensors to measure positional offsets and initiate a corrective movement (active methods).

Force problems
If the dimensions of the components are small, forces acting on their surfaces (for instance coulomb or adhesion forces) may exceed their mass inertia. The resulting interference forces may cause the part to behave in an uncontrolled manner. In contrast of this, the effects of surface forces can also be used deliberately in the handling of small components (e.g. adhesive gripper). A further problem resulting from the forces exerted is the high risk of damage because components are in many  cases highly sensitive.
The force problems mainly occur during gripping and joining operations. On account of their tactile operating principle the compliant systems can only be used if the contact forces needed for tolerance compensation have been minimised. Forces can be minimised by using design elements that are free from friction and stick-slip (e.g. air bearings or solid-state joints) for assembly components. Sensors can be integrated into the assembly modules to monitor the gripping and assembly forces.
 
Interference factors
Vibration, temperature changes or contamination can lead to positioning errors or have an adverse effect on product quality. In addition to external (environmental) influences, internal sources of interference, which belong to the plant, also have to be considered (e.g. vibration and heat caused by drivelines, bearing wear etc..). the interference problem can be resolved by avoiding variable influences (e.g. working in a clean room), and compensating for influencing effects (e.g. for vibration and displacement caused by changes in temperature).

Variety of models
A notable feature of miniaturised products is their large number of different models. An assembly system can be adjusted to various models either by a modular design which uses easily interchangeable product-specific system components (external flexibility), or by the processing of information (internal flexibility).

Gripper problem
One particularly major problem is posed by gripper technology: in some cases, the structure of the components involves a great deal of filigree work, and the gripping must not be allowed to destroy them or impair their function. Like-wise, the large amount of space needed to install the gripper must not be allowed to impede the joining operation. The following principles, which are applied and examined in general assembly, may be summarised in relation to gripping for microsystem engineering components:
· non-positive gripping by means of form elements
· positive gripping by means of form elements
· gripping by means of a vacuum
· gripping with adhesive substances
· magnetism
· electrostatics
· piezoelectric effect
· form memory effect

References

[Nof85]  Shimon Y. Nof, Handbook of Industrial Robotics, Wiley 1985, ISBN 0-471-89684-5.

[Gro86]  Mikell P. Groover, M. Weiss, Roger N. Nagel, Nicholas G. Odrey, Industrial Robotics, McGraw-Hill
              1986, ISBN 0-07-024989-X.

[And96] Anders R. Ericsson, Development and Implementation of the FACE Control Module for Flexible Automatic
              Assembly, Licentiate Thesis, The Royal Institute of Technology, Stockholm 1996.

[Ang97] Dr Marcelo Ang, Active Compliance Control of a PUMA 560 Robot, Research News Vol. 12 N°3, National
             University of Singapore, http://www.eng.nus.sg/EResnews/May97/may97p10.html.

[Sch97] Dr. sc. techn. Ullrich Schmucker, Fraunhofer-Institut IFF Automation - Project: Intelligent Six Components
             Force Torque Sensor with Integrated Signal Processing, http://www.iff.fhg.de/iff/aut/projects/kraft/kraft_e.htm.

[Hor93] K. Horikami, Y. Itsuzaki, M. Nakao, M. Nakamura, M. Takano, K. Okumura, Vision for Screw Fastener
             Robot, FA Engineering Laboratory, Matsushita Electric Industrial Co., Ltd., 24th International Symposium on
             industrial Robots, November 4-6 1993, Tokyo, Japan.

[Bad93] F. Badano, M. Betemps, C.W. Burckhardt, R. Clavel, A. Jutard, Assembly of Chamferless Parts Using a Fast
             Robot, Industrial Automation Laboratory, National Institute Of Applied Sciences, Villeurbanne France, Institute
             Of Microengineering, Swiss Federal Institute of Technology, Lausanne Switzerland, 24th International
             Symposium on industrial Robots, November 4-6 1993, Tokyo, Japan.

[Pau94] E. Paulos, J. Canny, Accurate Insertion Strategy Using Simple Optical Sensors, Department of Electrical
             Engineering and Computer Science, University of California - Berkeley, June 1994, Internet:
             http://vive.cs.berkeley.edu/~paulos/peg_in_hole.html

[Rei97] G. Reihnhart, M. Höhn, Growth into Miniaturisation - Flexible Microassembly Automation, Annals of the CIRP
            Vol. 46/1/97.
 
 Back to Home Page