Methodologies for the Integrated Modeling and Optimization of Digital Supply Chains

This thesis addresses the need for a more integrated approach to supply chain management. Such an approach must consider the relationships between the different flows in the supply network: information, physical, and financial flows. The thesis is organized in three parts. Part i focuses on the modeling of information flows in the supply chain, as captured in business processes such as the order-to-cash and source-to-pay processes, wherein a request travels through a network of transactions that modify the state of the request until it is fulfilled. Various chemical plant scheduling models from the process systems engineering (pse) literature are extended to model the flow of customer orders through a business process network, making the analogy that a business process is like a multi-purpose chemical plant and an order is like a batch of product. Of the models extended, the discrete-time state task network (stn) is found to perform best in larger problem instances due to its continuous relaxation tightness and its amenability to performance boosts by commercial solvers. A digital twin framework for business processes is then presented, which leverages discrete event simulation (des) and discrete-time stn models for online scheduling of customer orders in an order-to-cash process network under demand and processing time uncertainty. The use of des as a tool for uncertainty propagation in disturbance impact forecasting and order fulfillment date estimation is also discussed. Part ii turns to the physical processes in the supply chain by first comparing four network-based discrete-time scheduling models used in chemical batch scheduling operations: stn, mstn, rtn, and uopss. These models are also extended to include quality based changeovers (qbc), an important type of changeover event that has not been previously considered in the pse literature. An initial comparison shows that the rtn model is the most computationally efficient for large problem instances, whereas the uopss model has a more intuitive representation and a performance that is in between that of the stn/mstn and rtn models. In addition to this study on material transformation processes, the storage and distribution of materials in the supply chain is studied from the perspective of dynamic inventory management. Three approaches to dynamic inventory replenishment decision-making are compared and validated in a stochastic discrete-time simulation: deterministic linear programming, multi-stage stochastic programming, and reinforcement learning. The stochastic programming approach yields the best performance in terms of profit, whereas the reinforcement learning approach is more conservative and distributes the inventory throughout the network.

Part iii presents models that integrate the different flows and processes studied in parts i and ii. A model is presented for integrating the order-to-cash process with a make-to-order batch chemical plant model. The model performance is evaluated in an extended version of the digital twin presented in part i, which simulates both transactional and physical processes. The integrated model is then compared against single-focused models that either ignore the transactional network or simplify the physical processes involved. The integrated model shows the benefits of capturing the details of both the material and information flows, as well as the relationships between them, providing schedules that result in higher system profit relative to the single-focused models. In order to extend this model to include more than one supply chain echelon and financial flows, another model based on the stn and rtn modeling approaches is proposed. This model has the objective of maximizing shareholder equity while modeling inventory replenishment decisions, chemical production scheduling, business process operations, and financial assets and liabilities. This model is followed by a study of the generalized disjunctive programming (gdp) modeling framework as a promising modeling approach for modeling integrated supply chains. The gdp model is extended to capture higher order logic and explicitly represent hierarchical decision levels in the supply chain.

With the different models and methodologies presented, this thesis lays the groundwork and paves the road towards the development of the next-generation decision-support systems for a digital supply chain.