Reliability improvement of a soft drink production line using a Bayesian network
P.M.H. Freling (TU Delft - Mechanical Engineering)
R.R. Negenborn – Mentor (TU Delft - Mechanical Engineering)
Y. Pang – Mentor (TU Delft - Mechanical Engineering)
G. Monchen – Mentor
B. Zwerink – Mentor
M. Borsotti – Graduation committee member (TU Delft - Mechanical Engineering)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Efficient maintenance scheduling is critical to ensure the reliability and cost-effectiveness of production lines. Traditional approaches often rely on corrective maintenance and limited preventive maintenance, which can lead to unplanned downtime and higher operational costs. This research investigates the impact of predictive maintenance, supported by a condition-based model, on the reliability and cost performance of a production line. Due to insufficient real-world condition monitoring data, a synthetic dataset was generated based on historical process data and literature assumptions. This dataset simulates the running time, probability of cavitation, and operational load conditions for valves and pumps across the production line, allowing the model to provide failure probability estimates and recommend maintenance interventions before production begins.
The study compares two scenarios: the first follows conventional maintenance practices with primarily corrective actions and minimal preventive maintenance during scheduled production-free weeks; the second scenario integrates the predictive maintenance model to guide both preventive and predictive interventions. Key performance indicators (KPIs) are defined to evaluate the effect of the model on line reliability and maintenance costs. The primary KPI, the Maintenance Downtime Index (MDI), measures the ratio of planned maintenance hours to total downtime hours, reflecting the proportion of downtime that is scheduled versus unplanned. Additional KPIs analyze the distribution of maintenance costs among corrective, preventive, and predictive actions, with higher proportions of predictive maintenance indicating improved reliability.
Results demonstrate significant benefits of using the predictive maintenance model. The MDI shows a reduction of 5% in total downtime hours due to fewer unplanned interruptions and a greater allocation of downtime to planned maintenance activities. Maintenance cost analysis reveals a 53% reduction in total costs when predictive maintenance is applied. Furthermore, the proportion of corrective maintenance costs decreases substantially, confirming that the model effectively shifts maintenance efforts from reactive to proactive interventions. These findings indicate that predictive maintenance enhances both operational reliability and cost efficiency, supporting more informed decision-making by operators and maintenance planners.
The study highlights the importance of integrating condition-based predictive models into production scheduling, even when limited real-time data is available. Synthetic datasets, grounded in historical data and validated assumptions, provide a viable approach to evaluating predictive maintenance strategies and their impact on key operational metrics. By prioritizing predictive interventions over corrective actions, production lines can achieve lower downtime, improved reliability, and reduced maintenance expenditure. The findings offer practical guidance for manufacturing operators seeking to optimize maintenance strategies and support the broader adoption of predictive maintenance in industrial settings.