A synthetic multivariate refrigerator time-series dataset for predictive maintenance

This dataset contains synthetic multivariate time series generated for 27 refrigerators using a simplified physics-inspired simulator at one-minute resolution. Each refrigerator is simulated under its own configured operating and usage conditions, including ambient conditions, door-use patterns, thermostat settings, electrical characteristics, defrost schedules, degradation settings, and sensor-output configuration. The simulator represents several operating phenomena, including ambient-temperature variation, door use, thermostat and compressor operation, heat exchange, defrost, electrical consumption, and six types of progressive degradation. Depending on its configuration, each refrigerator provides 15 to 20 sensor outputs. The released data include 27 time-series files and 27 corresponding failure logs with explicit degradation start times, failure occurrence times, and failure end times. The dataset can be used for different predictive-maintenance tasks, including failure prediction, degradation detection, remaining useful life estimation, anomaly detection, and early-warning analysis. The diversity of operating conditions, usage patterns, and sensor-output configurations also supports studies involving heterogeneous equipment and learning across refrigerators. The Python data generator and refrigerator configuration files are provided as well, allowing users to generate additional synthetic data under different operating conditions, usage patterns, degradation settings, failure schedules, and sensor-output configurations.