Construction machinery is a determinant factor in special civil engineering. Efficient use of the available machines is important both in terms of efficiency and costs, as well as from an environmental perspective. The foundation for any improvement is the correct detection of the actual state and the currently performed activities and their duration. Existing approaches to automatic activity recognition most often require additional attached sensors, which is not always feasible in special civil engineering. On the other hand, certain machines employed in special civil engineering already possess their own machine data, which are collected and transmitted. This thesis addresses the automatic activity recognition of a Kelly drilling rig based on available machine data using Deep Learning methods. The proposed approach is based on a hybrid model, which consists of a combination of convolutional and uni- or bidirectional recurrent layers. The hybrid model is compared with a feedforward neural network and a LSTM network as baseline models. Furthermore, the influence of sample width, sample overlap and the possibility of performing a hierarchical classification are investigated. The use of the hybrid model fundamentally demonstrates good results compared to the two baseline models. Furthermore, the implementation of a hierarchical classification showed to be in part advantageous. In this case, the selection of the labels used could also be identified as a problem.
«
Construction machinery is a determinant factor in special civil engineering. Efficient use of the available machines is important both in terms of efficiency and costs, as well as from an environmental perspective. The foundation for any improvement is the correct detection of the actual state and the currently performed activities and their duration. Existing approaches to automatic activity recognition most often require additional attached sensors, which is not always feasible in special civi...
»