Equipment data-based activity recognition of construction machinery

Heavy equipment plays an essential role in construction projects. However, to optimize the allocation of that resource (e.g., via discrete-event simulation), the correct monitoring of the current usage of the equipment is imperative. Machine learning methods are usually based on the mounting of additional sensors such as accelerometers or gyroscopes on the equipment. However, many kinds of equipment already transmit production data to a proprietary platform. This paper explores the possibilities of performing activity recognition based on already available equipment data transmitted by the equipment, without the need for additional sensors. The proposed approach is based on hybrid models consisting of convolutional neural networks (CNN) and recurrent neural networks (RNN). Additionally, the influence of the use of bidirectional RNN is investigated. The hybrid models are compared with other baseline models. Both hybrid models demonstrate good results compared to the baseline models. Furthermore, the influence of the sample size is examined.