(54al) Multi-Objective Stochastic Optimization for Preventive Maintenance Planning in Chemical Plants

Maintenance planning and process operations in chemical manufacturing facilities are subject to several sources of uncertainty ranging from volatile feedstock prices to uncertainty in the level of demand. In the context of assuring the integrity of assets in ageing chemical plants, the present research focuses on uncertainty in equipment availability and develops a novel multi-objective stochastic mixed-integer nonlinear optimization algorithm for preventive maintenance planning. The algorithm factors in uncertainty to arrive at robust optimal solutions in contrast with other approaches such as risk-based inspection and deterministic optimization. In addition, it simultaneously considers the two competing objectives of cost minimization and system reliability maximization to decide on the optimal maintenance frequency, set of online equipment and process flowrates at each time stage. The proposed approach consists of two main steps. In the first step, the system is represented as a dynamic Bayesian network to capture complex interactions between system components, identify critical failure pathways, and characterize the overall risk of system failure. In the second step, the multistage stochastic optimization subproblem is formulated over a shrinking time horizon to progressively incorporate decisions and information from prior time stages into the decision-making process. Following system representation and problem formulation, the overall multi-objective optimization problem is solved using the epsilon-constrained method to obtain the optimal maintenance policy. The overall optimization problem is non-convex and large-scale and computational difficulties are tackled within the algorithm using structure-based decomposition strategies such as Generalized Benders decomposition, or via metaheuristic optimization techniques such as simulated annealing. The results of the research include: (i) obtaining a robust optimal expected preventive maintenance plan and operating schedule, and (ii) provision of a Pareto front of optimal solutions from which the decision maker can select. This robust approach combining the techniques of Bayesian network-based risk assessment and nonlinear stochastic programming is illustrated with a case study and can be used to improve overall equipment availability and maximize plant productivity.