Dzenan Jupic

Dzenan Jupic

Werner-Heisenberg-Gymnasium

 

Titel der Forschungsarbeit: Schüttguterkennung mittels neuronaler Netze

Fakultät: Fakultät für Maschinenwesen

Lehrstuhl: Lehrstuhl für Fördertechnik Materialfluss Logistik

Betreuung: Maximilian Schöberl

Abstract der Forschungsarbeit

The goal of this research work was to find out, if it is possible to create a system based on artificial intelligence, that allows distinguishing different types of soil, transported by a wheel loader, confidently. The hypothesis was, that convolutional neural networks pared with a good camera would do the job.

The used setup consisted of the Microsoft Kinect 2.0, a computer and a voltage converter on the hardware side, TensorFlow, Python and C# on the software side.

For collecting training data, the camera was mounted onto the roof of the wheel loader, while the Computer and the voltage converter were placed inside the cabinet.  The whole setup was tested under different conditions at the two companies Wacker Neuson SE and H. Geiger GmbH.

In the end, it turned out that both CNNs developed achieved accuracies of about 90%. These results show that it is possible to distinguish different types of soil with an ai based system.