Data-Driven Defrost-on-Demand Scheduling in Reefer Ships: A Predictive Maintenance Framework Using Real-World Sensor Data

Abstract

Defrost cycles in shipboard refrigeration plants are typically initiated on fixed schedules or by operator judgement, which can lead to unnecessary energy use and temperature variability during the transport of frozen products. This study proposes a data-driven predictive maintenance approach to trigger defrost on demand in a merchant reefer vessel carrying frozen tuna. A total of 76,692 operational records from cargo-hold air coolers and holds were analysed, including delivery and return air temperatures, hold air temperature and relative humidity. The records were modelled to show their behaviour under real voyage conditions. The modelled strategy indicates that 66.55% of defrost cycles performed during the study period were unnecessary, suggesting substantial scope to reduce defrost frequency and associated disturbances. These findings demonstrate the feasibility of implementing machine learning (ML)-based decision support for maritime refrigeration. This enables defrost-on-demand scheduling, which has the potential to enhance operational efficiency while supporting product quality, sustainability and traceability in the frozen tuna supply chain.