Swire-Cola-Capstone-
Predictive Maintenance model for Swire Coca-Cola
Overview
This project focuses on developing a predictive maintenance solution for Swire Coca-Cola to minimize unplanned machine downtimes. Leveraging survival analysis and machine learning models, the solution aims to forecast machinery failures, enabling proactive repairs, cost savings, and enhanced operational efficiency.
Swire Coca-Cola’s production plants face significant challenges with unplanned equipment breakdowns, resulting in production shortfalls and an estimated $60 million annual loss. The current reactive maintenance strategy leads to delayed repairs, increased costs, and reduced production efficiency.
To address these challenges, a predictive maintenance system was developed. This system incorporates:
- Survival Analysis: To estimate the likelihood of machine failures over time and predict high-risk periods.
- Random Forest Models: To identify key features influencing breakdowns and generate highly accurate predictions.
- Logistic Regression: To predict probabilities of equipment failure under specific conditions.
These models provide actionable insights to optimize maintenance schedules, reduce costs, and improve production continuity.
Business Value
This solution delivers:
- Proactive Maintenance Planning:
- Reduces unplanned downtimes by identifying high-risk machinery.
- Improves repair efficiency and resource allocation.
- Operational Efficiency:
- Keeps production running smoothly, reducing disruptions.
- Meets customer demands through consistent production schedules.
- Cost Savings:
- Aims to cut downtime-related losses by at least 50%, saving Swire Coca-Cola millions annually.
Challenges
The project encountered several challenges:
- Over 80% of missing equipment IDs, reducing model granularity at the machine level.
- Imbalanced breakdown data, requiring careful handling to avoid overfitting.
- Inconsistent date fields, necessitating extensive preprocessing for accurate analysis.
Contributions
This project provided an opportunity to:
- Handle real-world messy data, learning essential preprocessing techniques.
- Conduct a random forest model to see seasonal effects of broken-down machines.
- Frame and execute research to align with business goals while balancing technical rigor and stakeholder concerns.
Additional Resources