Cognitive Assembly Cell

  Robot Cluster of Excellence Integrative Production Technology for High-Wage Countries

The degree and the scope of automation of the technology used increased with the growing number of quantities. One reason was the limited flexibility of the automated systems. The effort to design, program and operate such a system safely were high in the case of frequent changes in the production process. Today‘s automation systems are characterized by rigid rules and have no intelligence nor the skills to develop intelligence. An automation technology that limits the variety of products and the flexibility of production processes also hinders the individualization of production. The ability to adapt to changing demands and environmental conditions is therefore
a central requirement for future automation systems. However, the automatic decision-making requires a much broader knowledge within the automation systems. This expansion of knowledge demands an increase of the possible transparency of internal processes of the automation technology, so that in case of a problem the operator can, for example, also be involved in the solution process.


Practical Issues

Planning cycle CoE Planning cycle of the cognitive architecture

The increasing demands on the automation of processes within production lead to a disproportionate increase in planning and preparation activities. An increasing automation is only useful if the required adaptability can be compensated by a reduction of the necessary effort. These costs are in particular reasoned in the field of nonvalue added planning, design, programming, commissioning and testing. Even the close integration and networking of real automation technology with the virtually generated models opens up potential. Finally, the degree of automation in carrying and supporting of planning tasks can be increased further.



Cognitive control architecture Cluster of Excellence Integrative Production Technology for High-Wage Countries

The approach aims at shifting the planning process into the execution level. This will require an “intelligent” control platform capable of planning and executing the process autonomously with regard to the target and the current situation and to interact with the human expert. Thereby, the (given) generic structure is always distinguished from the (acquirable) problem specific content. The sum of both is finally observable as intelligent behaviour. The structure is also named cognitive architecture and has deep interdependencies with the representation of the contents. In the following application, a methodical approach for the integration of a cognitive control architecture into production systems is demonstrated using a robotic assembly cell: the operator of the assembly cell specifies an assembly task by composing simple basic components to a complex assembly group by means of a graphical interface. In order to realize high transparency, variability and scalability and to keep the complexity of the actual assembly process as low as possible, Lego- Duplo components have been used. The task of the cognitive control unit is to find a possible assembly sequence of the components at runtime, to plan the assembly autonomously with respect to the available resources and to control the process. The assembly cell comprises two robots and a transfer system consisting of six independently controllable conveyor belts with gates and light barriers. The assembly robot uses a camera for identifying the components and a flexible gripper for handling the components. Beside these active components, different areas which can be used for the assembly and buffering also exist. The assembly groups that are finished are then checked with regard to correctness of the geometry and colour by means of 3D image processing by the second robot.


Technical Challenges

Gripping process Cluster of Excellence Integrative Production Technology for High-Wage Countries

The most relevant technical challenges are the development of the algorithms necessary for planning and controlling the assembly process. These are planning algorithms used for generating the possible assembly sequences as well as technical algorithms used for instance for 2D and 3D image processing. The possible assembly sequences are determined by means of graph-based algorithms. Therefore, at first the possible assembly steps are determined using a so-called Assembly-by-Disassembly strategy. The assembly group is disassembled stepwise until the final component. The intermediate states describe in the reverse direction all the possibilities to perform the assembly. And afterwards, the resulting valid states are transformed into an assembly graph by combining equal states, and weighted for the further planning process. During the process of grasping an object, 2D image processing enables the identification of moving components. Due to the connection to the control unit of an industrial robot, their velocity can additionally be determined, so that the movement of the robot can be synchronised. By calculating the relative pose, the component can be grasped without any interception of the movement.
In order to verify the correct build-up, the colour and depth data are recorded from four sides. By combining these four images, the 3D representation of the assembly group is calculated. This representation is then compared with the predefined assembly group and deviations are communicated to the operator – this communication also has to be transparent. Furthermore, the variety of technical components and their interfaces are additional technical challenges. The decisions made in the cognitive control unit must be for instance, transformed into control commands for the robot and programmable controllers.