Object-Centric Process Mining (OCPM) requires discrete events that explicitly reference activities and the objects they affect. In industrial processes where physical materials move as batches from one station to another, such as in the chemical industry, behavior is recorded solely by low-level sensors as time-series data. Without activities and object identifiers being recorded, OCPM is inapplicable. This paper proposes a novel methodology to transform raw time-series sensor data into semantically meaningful, discrete events, and to infer objects and relations that conform to OCPM requirements. We apply this methodology to real-world sensor data from a polyethylene terephthalate (PET) chemical recycling process. Our results show that this transformation enables object-centric analysis of industrial processes, validated through expert feedback and alignment with real material flows.

pre-conference proceedings version.