GarimaDahuja/openenv-manufacturing-rl

Manufacturing Process Optimization (OpenEnv RL)

Overview

This project implements a reinforcement learning environment for manufacturing process optimization using OpenEnv. The environment simulates a factory where jobs must be assigned to machines efficiently to minimize delays and maximize utilization.

Motivation

Efficient scheduling is essential in manufacturing systems. This environment models real-world constraints such as machine availability, job priorities, and machine breakdowns, enabling the development of intelligent scheduling strategies.

Observation Space

The state includes:

• —machines: list of machines with id, status (idle, busy, broken), remaining_time, breakdown probability
• —jobqueue: list of jobs with id, processingtime, and priority
• —current_time: current timestep

Action Space

An action consists of:

• —machine_id: selected machine
• —job_id: selected job

The agent assigns a job to a machine.

Tasks

Three tasks are defined:

1. 1.Minimize Idle Time (Easy) Score = 1 - (idle machines / total machines)

1. 1.Optimize Completion Time (Medium) Score = optimaltime / actualtime

1. 1.Handle Machine Failures (Hard) Score = working machines / total machines

The overall score is the average of all three tasks.

Reward Function

The reward provides:

• —positive reward for job completion
• —penalty for idle machines
• —penalty for delay based on time

Rewards are normalized between 0 and 1.

Setup

Install dependencies:

pip install -r requirements.txt

Set environment variable:

set HF_TOKEN=dummy

Run inference:

python inference.py

Docker

Build:

docker build -t openenv-rl .

Run:

docker run openenv-rl

Baseline Performance

Typical results using the baseline agent:

• —Task 1: 1.0
• —Task 2: 1.0
• —Task 3: 1.0
• —Overall: 1.0

Scores may vary due to randomness.

Project Structure

env/
  env.py
  models.py
  tasks.py
  openenv.yaml

agent.py
inference.py
test_env.py
requirements.txt
Dockerfile
README.md