IoT Sensor and Maintenance Dataset for Machine Learning–Based Availability Analysis in Food Manufacturin
Description
The research hypothesis underlying this dataset is that the integration of machine learning techniques with IoT-based monitoring and Total Productive Maintenance (TPM) strategies can significantly improve asset availability and reduce unplanned downtime in food manufacturing systems. The dataset shows the temporal behavior of a cooling tunnel under different operational and maintenance scenarios, including baseline conditions and improved strategies incorporating predictive analytics. The data capture failure occurrences, downtime durations, repair processes, maintenance intervals, and system performance indicators derived from multiple discrete-event simulation runs. The results highlight that predictive and condition-based maintenance strategies lead to higher availability levels, reduced variability in system performance, and improved utilization of critical resources when compared to traditional maintenance approaches. These findings suggest that data-driven maintenance policies enable more effective decision-making, particularly in environments with high operational variability. The data were collected through multiple independent simulation replications developed in Simio Simulation, ensuring statistical validity and reproducibility. Researchers can interpret and reuse this dataset to train and validate machine learning models, evaluate maintenance policies, perform reliability and availability analyses, and develop decision-support tools applicable to manufacturing systems, especially within food industry contexts.
Files
Steps to reproduce
The data were generated using a discrete-event simulation model developed in Simio Simulation (Academic Version) to represent the operational and maintenance behavior of a cooling tunnel in a food manufacturing production line. The model incorporates process flows, resources, failure and repair events, and maintenance activities using an object-oriented approach. Model input parameters, including processing times, interarrival times, failure rates, and repair times, were defined using statistically fitted distributions based on historical operational data. Multiple simulation runs were executed under different operational and maintenance scenarios, including a baseline scenario and improved maintenance strategies. Each scenario was evaluated through multiple independent replications to capture system variability and ensure statistical validity. The recorded outputs include time-based variables, failure events, downtime and repair durations, and aggregated performance indicators such as availability, cycle time, and resource utilization. The dataset can be reproduced by reconstructing the simulation model and executing equivalent replications in Simio Simulation.
Institutions
Categories
Funders
- Universidad Peruana de Ciencias Aplicadas