Hybrid Flow Shop Solver (HFS-SDST)

A production scheduling application for the Hybrid Flow Shop problem with sequence-dependent setup times (SDST) and learning effects.

About the Project

The project combines:

• a JavaFX UI layer for instance definition and result presentation,
• a Python optimization engine,
• JSON-based data exchange through data/ and results/.

The main objective is to minimize $C_{max}$, the completion time of the last job (makespan).

Scheduling Problem (Brief)

The addressed problem is HFS-SDST:

• HFS (Hybrid Flow Shop): jobs pass through consecutive production stages, with multiple parallel machines available at each stage.
• SDST (Sequence-Dependent Setup Times): setup time depends on the order of jobs.
• Learning effect: processing times may decrease with the job position in the sequence.

In practice, this is a complex combinatorial optimization problem where schedule quality depends on both job sequencing and machine assignment.

Technology Stack

• Java 17 - desktop application and UI logic.
• JavaFX 21.0.1 - user interface and result visualization.
• Maven 3.6+ - Java module build system.
• Python 3.8+ - optimization engine.
• Matplotlib - Gantt chart generation.
• Pytest - unit testing for the Python module.

Repository Architecture

Hybrid-Flow-Shop-Solver/
├── start.bat
├── README.md
├── Demo_Ui/
│   ├── README.md
│   └── javafx-ui/
├── Python/
│   ├── README.md
│   ├── TESTING.md
│   ├── main.py
│   ├── algorithms/
│   ├── core/
│   ├── utils/
│   └── tests/
├── data/
│   └── README.md
└── results/

Module documentation:

• UI: Demo_Ui/README.md
• Python backend: Python/README.md
• Python tests: Python/TESTING.md
• Input data: data/README.md

Available Algorithms

• Greedy MSTF - fast baseline heuristic.
• Tabu Search - metaheuristic with a strong quality/time tradeoff.
• Branch and Bound - exact method, recommended for small instances.

Requirements

• Java 17+
• Maven 3.6+
• Python 3.8+

Quick Start (Windows)

From the project root:

start.bat

The startup script prepares the environment and launches the application.

Manual Run

1. Install Python dependencies:

cd Python
pip install -r requirements.txt
cd ..

2. Start the JavaFX interface:

cd Demo_Ui/javafx-ui
mvn clean javafx:run

Data Flow

1. The user defines a problem instance in the UI.
2. The UI writes input to data/input_<timestamp>.json.
3. The Python module runs optimization.
4. The result is written to results/result.json.
5. The UI presents metrics and a Gantt chart.

Input and Output Format

Input example (JSON):

{
  "algorithm": "tabu",
  "num_stages": 2,
  "num_jobs": 3,
  "machines_per_stage": [1, 2],
  "learning_coeff": 0.15,
  "learning_stages": "10",
  "processing_times": [...],
  "setup_times": [...]
}

Output example (JSON):

{
  "time_in_ms": 3123,
  "Algorithm": "Tabu Search",
  "C_max": 101.48,
  "gant_diagram": "<path-to-png>",
  "schedule": [...]
}

Practical Guidance

• Up to around 10 jobs: Branch and Bound can be considered.
• Medium instances: Tabu Search is usually the best compromise.
• Large instances: use Greedy for fast feasible solutions.

Testing

Unit test coverage is 94% for optimization-related code (Branch and Bound, Greedy, Tabu Search) and key domain classes (Job, Machine, Schedule).

Quick command:

cd Python
python -m pytest tests -v

Detailed testing and coverage guide: Python/TESTING.md.

Author

Bartłomiej Adam Kuk
Politechnika Wrocławska, Wydział Informatyki i Telekomunikacji

License

Project prepared as part of an engineering thesis.