Behavioral Changes Based Trust Computation in Mutli-Agent Predictive Maintenance Applications

Predictive maintenance practices rely on reliable scheduling and allocation of resources to sustain seamless operations. Advent of agent technology has increased the avenues to manage efficient schedules in distributed industry settings. This paper proposes a multi-agent architecture with behavior based trust computations and reinforcement learning to effectively manage multiple schedule assignments. To formalize the schedule maintenance problem, bigraph was used to model the centralized resource allocations among multiple schedules. Reinforcement learning algorithms was used to compute and update trust values based on the agent behaviors. Results reveal that the cooperation among agents helps to maintain a variation in assignments to less than 40%.