Every golfer is familiar with the situation where an unlucky ball does not land on the fairway or green, but in the nearest water hazard. Many thousands of golf balls, some of which have only been played once or a few times and may be in pristine condition, quickly accumulate there. Especially in the USA and Great Britain, where golf is a widespread sport and the density of golf courses is much higher than in Germany, vast quantities of balls lie dormant in the "lakes". Professional divers are out and about on golf courses around the world to retrieve these balls. The Wiesbaden-based company Easy Lakeballs specializes in collecting, cleaning and sorting lake balls and then putting them back on sale. Up to 3.5 million balls are reused in this way. The balls are particularly popular with low-handicap players or beginners, as a ball always ends up in the woods or in a water hazard, especially during the first few rounds of golf, which can be costly in the long run.

Manual sorting process time-consuming

The sorting process by brand and then brand-specific types is a very labor-intensive process. In order to counteract the prevailing labor shortage in this area, Easy Lakeballs decided in 2022 to explore the market with the task of automatic sorting. In the end, Meprovision, a provider of special solutions in the field of optical inspection and identification from Dillingen in the Saarland, was found as a partner to implement the project from the recognition side.

Automation solution requires AI

After a very intensive study and lengthy consideration of feasibility, it was clear that this project could only be solved using project-specific AI (artificial intelligence). There was also the challenge of presenting the balls to the cameras in such a way that the corresponding features (brand or type information) would most likely be visible. Once the right product to implement the AI had been found in Zebra Aurora Vision Studio, it quickly became clear that Minitec, as a long-standing partner of Meprovision, would be awarded the contract to develop and build the necessary automation technology and mechanics. Together, a transport concept was developed that guides the balls along a rail with the help of a segment conveyor. Three 12 MP color cameras capture a total of nine images of each ball, which are then used to determine the individual result for each ball. Using this method, an AI was created on the basis of around 50,000 images, which were captured specifically for the project and trained in a complex process. The system is therefore able to recognize a total of around 20 brands and up to 30 different types per brand.

Multi-stage process

The sorting process takes place in two stages. First, the balls, the composition of which is completely unknown for each test order, are sorted by brand. After this sorting step has ensured the purity of the marks, the individual marks are then sorted by type, color or type mix. The system has a total of 32 sorting points plus an NIO outlet and is capable of processing up to 7,000 balls per hour. The balls are discharged from the conveyor section via a pipe system and collected in different sized containers. Thanks to an upstream ascending conveyor with a hopper volume of up to 15,000 balls, the system can operate autonomously for just over two hours.

Diversity of types as a challenge

The greatest difficulty in connection with the task is the variety of types even within one class. For example, there are up to ten different logo representations to distinguish the Nike trademark alone. These differ in size and in combination with other characters. Recognizing the logos with rule-based approaches is therefore unthinkable due to the diversity. An enormous amount of effort is also required to train the AI, as the differences within the classes and the fact that there are many balls with logos and logo additions that are not known are enormously challenging.

MiniTec design with numerous special features

MiniTec had to adapt the design and automation of the system to the special task at hand. For example, the conveyor belt has 88 bolted-on cams to separate the golf balls during transportation. As soon as a golf ball has been identified by manufacturer and type, it is blown into the corresponding container. Each blowing nozzle is controlled individually (32 valves). The rejection point for unrecognized golf balls is monitored by a sensor. This serves both to count the "bad" balls and as overflow protection if the box is full and the balls can no longer be removed. General handling and work safety have also been taken into consideration. The entire system can be moved by hand using the lifting rollers if required. Two sensor-monitored double-wing doors are integrated for maintenance and service work. The machine is also equipped with an access guard made of 5 mm transparent polycarbonate panes.

System saves time and money

For the customer, the use of the sorting system means enormous savings potential in the sorting process and thus a certain independence with regard to the required personnel resources. Meprovision will continue to look after the system, particularly with regard to maintaining and updating the AI. Together with MiniTec, a highly customized solution was created for a very specific application that offers the customer significant advantages.