Event abstraction enables transformation of low level events into higher level events making process mining (PM) on sensor data available. There are many approaches to event abstraction described in literature, however the main approaches include supervised or unsupervised techniques. We address the challenge of transforming low-level sensor data into high-level activities required for effective process mining, a task traditionally reliant on domain experts. By leveraging LLMs to automate the labelling process of sensor data clusters, we bridge the gap between raw data and process models. Motivated by a mining industry use case, we validated the effectiveness of LLMs in accurately labelling operational phases. Our LLM-generated labelling rules demonstrated high accuracy and interpretability, simplifying the understanding for domain experts. Additionally, we compared our LLM-based approach with a Decision Tree Classifier, highlighting the advantages of LLMs in generating simpler, more understandable labelling functions. Our work underscores the potential of advanced AI techniques to enhance the efficiency and accuracy of PM, contributing to the Industry 4.0 initiative.