Case Study: 3D Scanners Improve Robots' Cake Decorating Skills

Mnet 148841 Cake Robotics Listing Image
Tim Dykstra, Engineer, Concept SystemsTim Dykstra, Engineer, Concept Systems

Dawn Foods outside Atlanta, Georgia, makes and decorates cakes to sell in large supermarket chains. Responding to retailers’ requests for more intricate decoration on cakes usually required specific handwork. The company was also receiving complaints about the quality of some decorations. Dawn Foods called on Concept Systems to solve these problems while retaining their existing robots.

Dawn Foods’ cake decorating operation included two lines, each with two Fanuc robots.

As cakes came down the line, a camera took an image of each one from above. From that image the robots could know the diameter of the top of the cake and its position on the conveyer. This type of 2-D visioning has been common in factories for years.

The robots would then apply decorative frosting, such as a border or dollops around the top edge of the cake, as well as a limited selection of other decorations, such as carrots made of green and orange frosting.

But every cake is different

A 2-D vision system cannot accurately tell the robots the height of the cake. The robots had to simply assume the cake was some specified preprogrammed height. But cakes don’t come out the same every time. The true height of each varies. Variance of as little as an eighth of an inch would cause distortion in the carrot shape, for instance. In fact, customers were complaining that the carrot shapes often looked wrong.

One customer also was asking for borders or dollops around the bottom edge of the cake, but the 2-D image didn’t provide accurate data about the exact shape and diameter around the bottom of the cake. Rarely is it precisely the same as the top.

The company’s solution was to do the bottom borders by hand — a significant investment of time and labor.

Mnet 148843 Cake Robotics Inline Pic2

Changeover delays

Another problem had to do with constraints of the 2-D camera, which took a photograph in poor lighting. When the plant switched from vanilla to chocolate cake, the camera needed to be adjusted to account for changes in contrast. To ensure accurate images, operators had to adjust the lighting, do some calibration, take a sample image and modify the settings until things were looking good.

Along with vision system optimization, the operator or a maintenance person had to also change the program in the robot through a separate controller and interface. Throughout this time, the line was effectively down.

Difficulty Adding New Cake Designs

If Dawn Foods wanted to add new cake designs, the process was quite difficult and involved very specialized expertise that the plant didn’t have available onsite. A new program had to be created within the 2D vision system and robot controller for each new cake design. This required someone with experience programming robots, and sometimes it was necessary to create the program from scratch. Due to these complexities, Dawn rarely added new cake designs to the system.

Solution: Going 3-D

The existing robots in the plant were fine, so Concept Systems came up with a plan to retrofit them with a more advanced vision system that could know what the cake looked like in three dimensions. In order to accomplish this, Concept created a new program that used the vision data as offsets.

This setup simplified the entire system, as everything could be done through a single, friendlier user interface. As opposed to modifying the vision system and robot controller each time cake designs changed, the operators could now use a single operator interface to alert the system that something different was running.

Two Hermary SL1880 3-D laser scanners were installed on each line. Rather than taking a photographic image like the old system, they use laser triangulation to create what’s called a “point cloud,” showing the precise shape of the cake, including its exact height every five degrees around its circumference.

This three-dimensional image of each individual cake is passed to the robots so they can adjust not only to the actual height but also the nuances of shape within each cake.

Mnet 148844 Cake Robotics Inline Pic1

Better quality, less handwork

In the end, Dawn Foods was able to create higher quality and more intricate frosting designs using the robots. A good deal of manual work was eliminated — for instance the application of a bottom border of frosting.

“It really expanded what this plant could do and sell,” said Concept Systems engineer Tim Dykstra, who worked on the project. “Dawn Foods’ sales team is able to talk about these new capabilities, too, to help sell products.”

Also, the new user interface allows much faster and easier adjustment of designs. A new cake design could be implemented through the friendly operator interface, rather than requiring a skilled programmer of both vision systems and the robot. And finally, the color of the cake does not interfere with the lasers, so changeover time lost to calibrating the vision system was eliminated.

Dawn Foods earned back its investment in the project in about eight months.

“We came in far under budget because the system was so easy to operate. Our training costs were far less than expected. While on site, Concept Systems also reviewed other parts of our project and gave solid advice in areas where they were not specifically looking for additional work — this really proved they are in business for the best interest of their customers.” — Dave Wojtowicz, Dawn Foods factory manager

Concept Systems, headquartered in Albany, Oregon, is a certified member of the Control System Integrators Association (CSIA), For more information on Concept Systems, visit the company profile on the Industrial Automation Exchange.

More in Operations