Session: Controls 3

Paper Number: 111383

111383 - Real-Time Prediction of Efficient Operating Points in Quasi-Stationary Agricultural Processes With Hydraulic Implements 

Agriculture is a critical industry that relies on the usage of mobile machinery with high energy demand. Agricultural machinery needs to become more efficient and automated to tackle the growing demand for food, the shortage of qualified labor, and the cost pressure due to the increase in fuel prices.

Currently, efficiency optimization is limited within target settings which are set by the human operator. Manual adjustments of the target settings are a challenging task, since the operator does not know the optimization strategy of the internal control systems of the machines, and therefore agronomic degrees of freedom can not be fully exploited for efficiency optimization.

This includes the working speed, which is directly related to the power demand of the system. The specified target speed does thereby restrict the optimization space.

While some agricultural processes only require draft power, others depend on additional power sources. Compared to the use of mechanical PTO shafts to provide energy for power-intensive functions of agricultural implements, the use of hydraulics offers the possibility of variable adjustment of the transmitted power as it is independent of the engine speed.

In this paper, an algorithm for prediction of the best possible operating point is proposed. The algorithm is based on the adaptation of the working speed in quasi-stationary agricultural tasks in combination with the necessary adjustment of the respective hydraulic power requirements.

The prediction system is based on the real-time identification and modeling of ambient conditions such as traction, and soil resistance. Additionally, it features a neural network for state assessment and selection. Based on the measured system state, the range of possible operating points is evaluated under the given process restrictions. With this approach, only the interaction of the tractor and implement with the environment needs to me modeled. The machine itself is represented solely by the neural network, which is trained with recorded machine data.

Based on a given reward function, the best possible operating point is determined and set as the new target value. Simulations demonstrate that the system is able to propose advantageous operating points.

Presenting Author: Benjamin Kazenwadel Karlsruhe Institute of Technology (KIT)

Presenting Author Biography: Benjamin Kazenwadel studied mechanical engineering at the Karlsruhe Institute of Technology and the Instituto Tecnológico de Buenos Aires, with a focus on mobile machines and energy technology. Since 2021, he is a PhD student at the institute for mobile machinery at KIT, focusing on the automation and efficiency optimization of mobile machines.