Adaptive ε-greedy exploration for stable reconfiguration in next-gen aviation IMA systems

Area of Science:

• Aerospace Engineering
• Computer Science
• Artificial Intelligence

Background:

• Next-generation aviation Integrated Modular Avionics (IMA) systems utilize container technology for improved resource utilization and flexibility.
• Container-based IMA architectures increase system reconfiguration complexity, posing challenges for fault tolerance.
• Current manual and heuristic reconfiguration methods are insufficient for modern fault tolerance demands.

Purpose of the Study:

• To develop an efficient embedded container reconfiguration method for container-based IMA systems.
• To enhance the adaptability and fault tolerance of IMA systems under continuous faults.
• To address the limitations of existing reconfiguration strategies in complex, large-scale deployments.

Main Methods:

• Implementation of a Double Dueling Deep Q-Network with Adaptive Epsilon-Greedy Exploration (D3QNAE).
• Utilizing adaptive exploration strategies to generate stable reconfiguration policies in complex environments.
• Evaluating the method in large-scale deployment scenarios (500-task configurations).

Main Results:

• D3QNAE reduced the time to the first feasible solution by 34% compared to the D3QN baseline.
• Achieved a 15.6% higher maximum reward value, indicating more effective strategies.
• Demonstrated a 100% migration impact rate under continuous faults, ensuring system stability.

Conclusions:

• The proposed D3QNAE method significantly enhances fault tolerance in container-based IMA systems.
• This approach offers improved stability and reduced maintenance costs.
• D3QNAE provides a robust solution for complex reconfiguration challenges in aviation systems.