Quote of the day

A person who never made a mistake never tried anything new.

- Albert Einstein

Contact

Multi-Agent Reinforcement Learning for Robust Job-Shop Rescheduling under Disruptions

Open Industry & Manufacturing

Multi-Agent Reinforcement Learning for Robust Job-Shop Rescheduling under Disruptions

Multi-Agent Reinforcement Learning for Robust Job-Shop Rescheduling under Disruptions

Description

Reschedule production in real time under machine breakdowns and urgent orders while jointly optimizing operational performance and schedule stability.

Details

Context and Problem Statement

Offline production schedules quickly become obsolete after breakdowns or urgent orders. Aggressive replanning can destabilize operations, while overly conservative replanning sacrifices performance.

Research Question

What trade-off between performance and schedule stability can a learned multi-agent rescheduling policy achieve compared with dispatching rules and full re-optimization?

Proposed Approach

Model each machine as an agent, train a reinforcement-learning policy with a reward combining tardiness and deviation from the previous plan, connect it to a digital twin through industrial interoperability protocols, and compare against dispatching heuristics and exact solvers.

Expected Contribution

An explicit evaluation of the performance-stability trade-off and a decentralized architecture suitable for real manufacturing environments.

Expected Prototype

A digital-twin demonstrator showing updated schedules, magnitude of changes, and estimated delivery impact.

Datasets

Taillard and Demirkol benchmark instances plus execution and failure histories from a partner workshop.

Challenges

Combinatorial complexity, multi-agent non-stationarity, model-reality mismatch, operator acceptance, and MES integration.

Research Question

Can a learned multi-agent policy reduce delays while limiting schedule instability more effectively than classical priority rules and full re-optimization?

Innovation

Schedule stability is included explicitly in the learning objective, reflecting a real operational constraint often ignored in academic scheduling studies.

Expected Deliverable

An open simulation environment, a trained rescheduling policy, and a documented comparative study of performance-stability trade-offs.

Technologies

Multi-Agent Reinforcement Learning Digital Twin and OPC UA Combinatorial Optimization and Benchmark Solvers Discrete-Event Simulation MES and Industrial-System Integration

Required Skills

  • Operations Research and Scheduling
  • Multi-Agent Reinforcement Learning
  • Industrial Simulation
  • Programming and Systems Integration

Datasets

  • Taillard benchmark instances
  • Demirkol benchmark instances
  • Execution and failure histories from a partner workshop

Morocco & Africa Relevance

Moroccan industrial subcontractors often operate under volatile order patterns and tight delivery constraints, making stable online rescheduling directly relevant.

References

  1. A Theoretical Framework for Graph-based Digital Twins for Supply Chain Management and Optimization
  2. Graph Neural Networks in Supply Chain Analytics and Optimization

Project Information

Duration: 4 to 6 months
Academic Level: Engineering, Master's or PhD
Research Area: Industry & Manufacturing

Keywords

Job-Shop Scheduling Multi-Agent RL Digital Twin Rescheduling Manufacturing

Evaluation

Scientific novelty
Impact
Technical feasibility
Data availability
Prototype potential
Industrial relevance

Apply for this topic

To apply for this topic, please send your CV and portfolio to:

contact@yanecode.com
Back to Research