Design and Operation of Process Supply Chains under Uncertainty

Abstract

This thesis deals with the problems of design and operation of process supply chains. Process supply chains face many challenges due to volatile market conditions, production and transportation delays, and stiff market competition, which ultimately affect their profitability. Supply chain management (SCM) is the process of managing the flow of materials and information within supply chain to optimize the SC performance. SCM is carried out using a hierarchical decision-making framework, where the top most layer looks at network design and the bottom-most layer deals with scheduling day-to-day activities. In this research, the systems engineering principles are applied to devise an improved methodology for supply chain optimization (SCO).

First, we consider the design of supply chain in the presence of demand uncertainty. The representation of network topology plays an important role in deriving the optimal network design. In real practice, the shipping cost for transferring goods from one location to another is determined based on service time and quantity. More importantly, the cost associated with establishing a transportation linkage is relatively small for existing transportation infrastructure and can be changed if beneficial. The flexibility of changing the transportation routes is included in the network topology representation by the explicit inclusion of time limited transportation contract agreements. Further, the customer demand is volatile, and it is very difficult to predict accurately. To handle the demand uncertainty, a two-stage stochastic programming formulation is applied in the SC design approach.

Next, we consider the problem of handling uncertainty in SC planning by applying a system engineering control principle, robust model predictive control (MPC). The uncertainty in model parameters (yield) and demand are captured by stochastic programming. In this approach, the planning activities are represented by a hybrid model with decisions governed by logical conditions/rulesets. An MPC based rolling horizon control framework is used to schedule the planning activities, where the SC performance is expressed using a multi-criterion objective comprising customer service and economics. The uncertainty in demand and yield are propagated by two mechanisms - an open-loop approach, and an approximate closed-loop strategy.

Finally, we consider the problem of integration of SC planning and scheduling. Due to the use of different time scale models for planning and scheduling, the decision derived from the planning layer may result in infeasibility when those targets are implemented at the scheduling level, which ultimately affects the supply chain efficiency. To address this issue, we model tactical and operational planning activities using an integrated hybrid time modeling approach in which the first few planning periods are formulated using an operational planning model and the remaining time periods are modeled with a tactical planning model. The main rationale for formulating an integrated model is that customer demand forecast becomes less accurate for a future time, therefore making a detailed planning model unnecessary. A key benefit of using a hybrid modeling approach is that it avoids the problem of infeasibility encountered in the hierarchical decision framework, as well as the computational burden associated with the use of a detailed planning model over a long time horizon. We employ an MPC based rolling horizon framework as a tactical decision policy where the integrated model is used to predict the system behavior.