Entwicklung und Evaluation eines Wiegesystems für Forstkräne auf Basis von künstlichen neuronalen Netzen

In both log and chip logistics, important reference data for logistic purposes are often lacking, as they are usually completed with insufficiently accurate estimates. In order to obtain higher quality information on the moving timber quantities, optional crane scales can be mounted between the telescope and the grapple of the forwarder. However, this has a negative effect on the crane kinematics and manoeuvrability while at the same time machine productivity is reduced due to an interruption in the loading process necessary for measurement.In this paper, a data-based method is presented which allows dynamic weighing in a continuous loading process for modern forestry cranes without the need to install an additional hardware component on the machine. This allows a cost-effective and comprehensive application. In the course of this method, a loading cycle is automatically detected, and the loaded mass is estimated by means of an artificial neural network (ANN). Signals from sensors installed as standard on modern forwarders serve as input variables. The Long Short-Term Memory (LSTM) architecture for the neural network has proven itself for handling these time-based sensor data. Based on LSTM cells, an appropriate network was designed, trained and subsequently optimized. The test shows an average full-scale error of 1.5% per 1,000 kg for a single loading cycle. For a fully loaded forwarder, this results in a total mass error of less than 1.2%.