Recently, the company installed a new automated machining line at its headquarters factory in Nordborg, Denmark. Now, six six-axis robots from Kuka perform many tasks that used to be done manually. The robots move components weighing between 5 and 20 kilograms each. Automating this process reduces the physically demanding work for the team, allowing workers to take on more value-added tasks.

The new line is highly flexible. The robots can handle 100 different components without manual intervention.

“That was the major challenge in this project,” explains Olav K. Sjørslev, sales manager at Quality Robot Systems (QRS), a systems integrator in Vissenbjerg, Denmark. “The requirement for the robotic cells at Danfoss was fully automatic feeding of the components, a task that could only be solved with a 3D vision system, given the large number of variations.”

The alternative—feeding the blanks meticulously ordered and aligned—was not economically feasible. Instead, the parts arrive at the first robot station in standard boxes. The boxes are stacked on top of each other more or less indiscriminately, and with the layers of parts separated by paper sheets.

The highly flexible line can handle 100 different components without manual intervention. Photo courtesy Kuka

“Our experience with vision systems in the past was not particularly good,” recalls Sjørslev. “We looked into various systems over the past decade, but did not find a single 3D vision technology that met our needs.”

What works at a trade fair or in a manufacturer’s demonstration room is often not viable under real-world conditions, Sjørslev points out. To meet the challenges of object recognition, QRS benefited from its close partnership with Kuka. When it comes to vision systems and object recognition, Kuka often turns to Roboception in Munich, Germany.

Vision Guidance in 3D

Roboception’s RC Visard 3D stereo vision system was ideal for the handling application at Danfoss. This smart camera generates and simultaneously processes location-based data in real time, enabling robots to perform applications such as bin picking. On-board processing power eliminates the need for external image-processing computers.

Parts pass from cell to cell via a conveyor. Photo courtesy Kuka

Available in five versions, the RC Visard is suitable for a range of applications. Using self-motion estimation, these sensors determine their position with millimeter accuracy, even in vibration-prone environments. Comprehensive on-board software, which can be extended by the RC Reason software suite, improves application-specific functionality, while an intuitive user interface makes setup easy. Engineers do not require any experience or knowledge in industrial image processing.

“Using our SilhouetteMatch software, the RC Visard recognizes components on the basis of a template created either from CAD data or from data generated directly by the sensor,” explains Michael Suppa, Ph.D., CEO of Roboception.

Cybertech robots feed the machined parts to a washing system, sort them into containers, or load them into another small grinding machine. Photo courtesy Kuka

Integrating the vision system with the Kuka robots was quick and easy using a standardized web interface and graphical user interface. The vision system reliably detects the position and orientation of the flat, unordered objects on the paper sheets and provides the robot with the gripping points.

“A 2D vision system was quickly ruled out for this project,” explains Suppa. “The paper sheets are not always completely flat or the parts may be oily.”

Roboception’s RC Visard 3D stereo vision system was ideal for the handling application at Danfoss. Photo courtesy Kuka

In tests with other vision systems, the camera had sometimes misinterpreted impressions of the parts on the paper as actual workpieces. The Roboception software did not have this problem. It recognizes exactly which components have been supplied to the cell and transmits the gripping points to a KR Cybertech six-axis robot.

In subsequent cells, additional Cybertech robots feed the machined parts to a washing system, sort them into containers, or load them into another small grinding machine.

Exchange of Expertise

For those involved, the project was very much a case of jumping in at the deep end.

“Implementation of the object recognition was a new experience for us,” says Sjørslev. “When we venture to take on something like this, we only select partners [that] we are sure will support us in the long term.”

He has only positive things to report about the cooperation with Roboception and Kuka. “For this project, everyone involved willingly went the extra mile in terms of transfer of know-how and time invested,” he says.

The effort was worth the effort in many respects. Danfoss was satisfied, and QRS will continue to work with Roboception in the future.

“We learned a great deal from this job that will help us to realize other projects,” says Sjørslev. “What is the influence of light conditions? What interference can occur on the lens? How far away can parts still be recognized? We now have concrete answers to these questions.”

Danfoss, too, is pleased with the project. The automation system has increased output, stabilized quality, and improved safety. And, if the parts change or additional work is added to the line, the system will be able to accommodate the changes.

“By integrating other software modules, our hardware can solve a wide variety of tasks,” says Suppa. “Danfoss now has a variable 3D vision system that can be easily adapted to other tasks as required.”