Resource-constrained management of heterogeneous assets with stochastic deterioration

We consider a collection of heterogeneous assets which exhibit independent stochastic behavior, but are economically interdependent due to resource constraints on management decisions. We represent the collection of assets as a network of nonhomogeneous Markov decision processes linked by side constraints. To facilitate a procedure for obtaining nonrandomized decision policies, we express the resource limitations as integrated chance constraints and relax them into the objective function within a Lagrangian penalty term. We utilize subgradient optimization to establish upper bounds and a greedy randomized Lagrangian repair heuristic to obtain feasible solutions. We empirically validate the tightness of these a posteriori upper and lower bounds with computational experiments on a pair of applications. For pavement maintenance, we examine the effect that the reward structure of each constituent MDP has on the maintenance policy in the presence of budget constraints. For equipment replacement, we consider constraints on two resource types.