The Robotics & Automation Laboratory (RAL) of the Institute of Production Engineering at Tampere University of Technology works for future automation platforms for manufacturing and assembly industries, incl. e.g. assembly automation, new assembly technologies, breakthrough laser applications in manufacturing and ICT in factory automation for enabling flexible and rapidly reconfigurable production systems.The laboratory has long traditions in light assembly production for electronics and mechanical engineering industries and in particular in automated final assembly and back-end assembly operations. One measure of the success in automated parts handling and assembly is the yield of the production line. Machine vision and other sensor based parts handling systems pave the way to more agile and flexible production lines capable of mass customization and small batch sizes.

The handling of small, sub-millimetric parts differs very much from the handling of traditional macro sized ones. The parts are so small that gravity no longer is the dominant force acting on them. This leads to intresting effects as for example adhesion and electrostaticity have a greater influence: the part might stand on its edge on the tray, it can get stuck on the robot gripper dangling only from one finger or it jumps towards the handling tool from a distance. Machine vision has a key role in sensing what is going on in the extremely dynamic micro world during production processes.

Traditional vision methods, such as 2D recognition and positional measurement, carry only to a certain point. In microscopic views micro parts usually show their individual surface features and no part looks exactly like its counterpart. Recognition must work more with resemblances than with stiff geometrics or templates. Position measurement and gauging has to work in three dimensions as no presumptions about the pose can be done. The lighting has to be extremely fast to turn on and off so that capturing miniature events during fast motions is possible. Because of the short on-time the lighting also has to be very powerful in order to allow integration of a decent image in the detector. Finally, the efficiency of the light source has to be high so that the small parts would not be stressed with too much heat.

Mechanics, actuators and control used for producing handling motions face a tough job as the required repeatability and accuracy often is well better than 10 microns. In addition to motion end points this usually affects the whole motion envelope and nonlinearities in transmission, bearings and guides have to be compensated in real time. Up to a certain level (some tens of microns) there are more accurate - and more expensive - versions available of mechanical components, but below that software methods are usually the only alternative even to achieve a simple straight line motion.

Basic research - tested in the real world

Recent research topics in RAL, covering basic, applied and confidential contract research, have included production systems for miniature products, execution of simultaneous fast and accurate motions as well as industrial control and communications allowing higher production flexibility. We have simultaneously also been striving towards miniaturizing production systems. Small production plants can be transported near the end customer and require less expensive floor space.

To study the abovementioned effects and ways to compensate them in real RAL has developed a miniature Cartesian robot, Minirobot, with 4 dof, outer dimensions of 500 x 500 mm and capable to 5 micron mechanical repeatability with maximum velocity of 4 m/s. The vision system utilizes a video microscope placed through the robot wrist. It can directly see everything happening under the robot and give correctional commands to the control system. The combined repeatability becomes better than 2 microns. Overall control runs on a standard PC, motion tasks on two commercial DSP cards and vision tasks on a specially programmed DSP. The robot has several grippers to choose from according to the task. It even has a transparent vacuum gripper for handling the most delicate components under constant vision control.

Same ideas have also been used in a highly integrated gripper module, Iris, that can be connected to commercial robot arms. The gripper contains a Firewire camera between its two linearly moving tool axes. To the tool axes one can connect fingers, soldering tool, tin feeder, fluid dispenser, vacuum gripper or electrical measurement probes. The vision system is an integral part of the task control and it controls the robot in 6 dof. The tool change units are located 'after' the camera and tool axes and costs per tool are cut low as the same motion mechanics and vision system can be used in all tasks. More information about Iris can be found from MVN 9/2004.

Volumetric measurements have become important as accurate dispensing of smaller droplets of viscous fluids is needed more often. The problem is that dispensing parameters might work well at one time, but a slight change in the ambient temperature for example can generate problems in outputting correct volumes in droplets. Laser triangulation - IR and visible range - has been used in RAL for nanoliter-scale measurements and the method can also be used for feedback when controlling fluid dispensing parameters. The same method has also been successfully used for measuring plastic parts with complex shapes. Modern sensor setups also enable simultaneous measuring of 3-D shape and visual quality.

Education guarantees the continuity

The educational task of the university is covered in RAL by offering a broad range of courses starting from basic principles of robotics and factory automation to in-depth courses e.g. in assembly, robotics, factory information systems and machine vision. All courses are lectured in English for facilitating fluent international student exchange. More than 70 students attend our vision course every year to learn about vision components, lighting solutions and image processing methods and tools in vision systems. Students from mechanical, electrical, signal processing, computing science and industrial economics are equally interested in taking the course. Clearly machine vision is maturing as a science and becoming part of basic skills of the engineer of tomorrow. Thus, RAL has plenty of promising research scientist material available in the future and still wide uncharted research terrains to explore in the fields of factory automation, robotics and machine vision.

Contact Information: Professor Reijo Tuokko Robotics & Automation Laboratory Institute of Production Engineering Tampere University of Technology Phone: +358-3-3115 2313 Email: reijo.tuokko@tut.fi Research scientist Jani Uusitalo Phone: +358-3-3115 4488 Email: jani.uusitalo@tut.fi