Automation and optimization of agricultural soil tillage applying machine learning based on machine- and process sensor systems

Climate change and cost pressure lead to new environmental and economic challenges that increase the demand for innovative control systems to automate and optimize agricultural tasks. Automating speed control during power-intensive soil tillage can increase eciency and sustainability and counteract the lack of qualied personnel in agriculture. A survey was carried out focused on tillage by cultivating to obtain an overview of the challenges farmers face during their work, including their target preferences. Based on the obtained requirements for tillage by cultivating, a system was developed automating working depth control by online Lidar plane detection to ensure tillage quality and establish a basis for good plant growth. Automated speed control is realized based on an online-parameterized draft force and traction model combined with the usage of a neural network for fuel rate prediction. The network is trained oine and adaptable to the individual preferences of the farms and varying implements. Thereby, the operator can choose and customize optimization objectives such as fuel eciency, performance, or total cost. During the evaluation, the control system was tested with various objectives against a human driver and was able to perform optimization on the individual objective. Furthermore, the transferability of the system was demonstrated with the usage of another implement.