Machine Process Classification in Conventional Tunnelling Using Accelerometers

With the world's ever increasing population and rapid urbanization, tunnels and underground structures provide a important solution to modern infrastructure problems. With the growing demand for underground construction comes the need for advancements in safety, sustainability and efficiency. The digitalisation and automation of process documentation in Building Information Modelling (BIM) aims to achieve these goals. In conventional tunnelling, where multiple machines perform sequential processes independently or in coordination with other machines, accurate documentation is crucial for optimising efficiency and safety. The minute diagram, a key record of construction activities, is traditionally completed manually, leading to delays and inaccuracies. Automating this process can improve reliability and aid project management. As part of a broader research project, this study aims to improve automation in tunnelling, and achieve advancements in underground construction by using accelerometer sensors to classify machine activities based on vibration signals. Machine learning algorithms are trained to analyse and automate process classification for real-time documentation. Using features extracted from the raw data, the trained models are able to accurately identify the various processes associated with a drill-jumbo. Confusion matrices and calculated performance metrics are used to analyse and compare the classification models.