LeMur Scheduler: An Innovative Approach to Spiral Binding Planning

🌀 LeMur Scheduler: An Innovative Approach to Spiral Binding Planning

Welcome to the project page of Team 3 - LeMur, part of the Industrial AI Challenge! Our mission? To revolutionize the way LeMur schedules production in its spiral binding department through intelligent automation and optimization.

📚 Project Resources & Demo

🔍 The Challenge

Efficient workforce scheduling and workload management are critical for any modern manufacturing process. LeMur sought a system that could generate optimized production plans, balancing workload distribution while respecting the unique constraints of their spiral binding operations—where manual interventions and asynchronous machine operations add complexity.

🎯 Project Goals

We set out to design a smart scheduler capable of:

• Suggesting optimal machine and operator schedules.
• Ensuring a balanced distribution of manual tasks.
• Delivering scalable and flexible planning over an extended time horizon—going beyond the company’s current day-by-day planning approach.

🧠 Problem Definition

To build an effective solution, we first developed a deep understanding of LeMur’s internal processes, defining key concepts like:

• Articles (products),
• Cycles (Setup, Load, Running, Unload, Machine Waiting),
• Operators and their roles.

We identified hard constraints (e.g., max simultaneous manual operations) and incorporated company-specific logic such as active/inactive cycles, partial/complete cycles, and machine capacity variations.

🛠 Methodology & Tools

The core scheduling problem falls under the Job Shop Problem (JSP) category, which is computationally complex. We chose a hybrid approach combining:

• ✅ Constraint Programming (for solution correctness and expressiveness),
• 🔁 Genetic Algorithms (for refinement and computational scalability).

We used:

• Google OR-Tools for CP modeling and solving,
• Inspyred for the genetic refinement phase.

Our 3-stage optimization pipeline:

1. Makespan Minimization
2. Compactness Maximization
3. Genetic Refinement

📊 Data & Synthetic Generation

Although historical data wasn’t used directly for learning, it helped us understand production patterns and design a synthetic dataset generator for testing. Input data includes:

• CSV files with ongoing and new orders,
• JSON files specifying articles and machine capabilities.

🗓 Weekly Development Process

Over several weeks, we followed a structured development cycle involving:

• Process analysis,
• Constraint formulation,
• Solver implementation,
• Continuous testing and validation,
  with regular feedback from LeMur’s mentors and domain experts.

✅ Results & Evaluation

Our hybrid solver delivers:

• Valid schedules within short computation times, even on complex scenarios.
• Progressive improvements through optimization stages.
• A strong match between generated and ideal workloads, often outperforming LeMur’s current planning system in terms of balance and load distribution.

⚙️ Performance Highlights

• Quick generation of first valid solutions
• Scalable on more powerful hardware
• Balanced operator workloads, minimizing bottlenecks and overloads

📊 Real-world Comparison

Tests showed our scheduler better reflects actual workload realities, highlighting areas where the current planning might underestimate operational demands.

🌐 Integrations & Tools

To make the system user-friendly and practical:

• A local web app was built for input management and Gantt chart visualization.
• A data converter was created to seamlessly interface with LeMur’s current planning formats.

🏭 Impact & Future Directions

The system paves the way for:

• Higher operational efficiency
• Better long-term planning
• Healthier workloads and improved work environments

Proposed future enhancements include:

• “Ghost production” suggestions based on demand forecasting
• Energy consumption and machine speed optimization
• Integration with LeMur’s ERP for full automation

💡 Conclusion

Team 3 proudly delivered a cutting-edge scheduling system tailored for LeMur’s spiral binding department. By combining Constraint Programming with Genetic Algorithms, we created a scalable, intelligent, and flexible solution. Validated by LeMur themselves, this tool stands to transform their production planning and workforce management for the better.