Process Mining of Sensor Data for Predictive Process Monitoring: A HACCP-Guided Pasteurization Study Case

Abstract

Industrial processes governed by food safety regulations, such as high-temperature short-time (HTST) pasteurization, rely on continuous sensor monitoring to ensure compliance with standards like Hazard Analysis and Critical Control Points (HACCP). However, extracting actionable process insights from raw sensor data remains a non-trivial task, largely due to the continuous, multivariate, and often high-frequency characteristics of the signals, which can obscure clear activity boundaries and introduce significant variability in temporal patterns. This paper proposes a process mining framework to extract activity-based representations from multivariate sensor data in a pasteurization scenario. By modelling temperature, pH, conductivity, viscosity, turbidity, flow, and pressure signals, the approach segments continuous data into discrete operational phases and generates event logs aligned with domain semantics. Unsupervised learning techniques, including Hidden Markov Models (HMMs), are used to infer latent process stages, while domain knowledge guides their interpretation in accordance with critical control points (CCPs). The extracted models support conformance checking against HACCP-based procedures and enable predictive process-monitoring tasks such as next-activity prediction and remaining time estimation. Experimental results on synthetic (literature-grounded data) demonstrated the method’s ability to enhance safety, compliance, and operational efficiency. This study illustrates how integrating process mining with regulatory principles can bridge the gap between continuous sensor streams and structured process analysis in food manufacturing.