•   When: Monday, July 30, 2018 from 10:00 AM to 12:00 PM
•   Speakers: Mohan Krishnamoorthy
•   Location: ENGR 1605
•   Export to iCal

This research is driven by the need for analysis and optimization solutions for production planning of complex service networks. Service networks are typically composed of unit manufacturing processes, base contract services, supply services, transportation services, and other supply chain components. These networks often involve physical or virtual inventories of products, parts, and materials that are used to anticipate uncertainties in supply and throughput of machines. These networks may be in steady state or may be temporal where the state of the processes, services, flows, and inventories changes over a time horizon until process completion. Additionally, the processes and services within service networks may be stochastic due to the presence of noise. This dissertation provides models, algorithms, and applications to compose service networks and perform different analysis and optimization tasks on them, and focuses on stochastic optimization algorithms based on white-box deterministic approximations.

The contributions of this dissertation are in three main thrusts: models, algorithms, and applications to service networks. More specifically, the contributions of this dissertation are given below. First, considered are service networks described over a time horizon and composed of inventories and general atomic processes modeled as piecewise linear functions. For these service networks, proposed is a modular modeling framework, which is used to develop deterministic and stochastic models to minimize the deterministic or stochastic cost of service network subject to deterministically or stochastically satisfying the demand with a given probability. Second, for the deterministic optimization problem, developed are manual translation methods from the modular modeling framework to mathematical programming formulation, which allows to apply mixed-integer linear programming algorithms to solve this problem. For the stochastic optimization problem, a special algorithm based on deterministic approximations is developed. Third, the developed modular models and algorithms are used in a collaborative research effort to build a prototype system to solve the deterministic and stochastic optimization problem over these service networks in a modular, extensible, and reusable way that is demonstrated using a real-world case study.

Fourth, deterministic and stochastic steady state models are developed for service networks composed of arbitrary physics-based models of unit manufacturing processes (UMPs) and other supply chain components. Fifth, the deterministic and stochastic steady state service network models are extended to include inventories and described over a time horizon. Additionally, composition and optimization of the service networks developed in the above two contributions are demonstrated using a case study.

Sixth, algorithms based on deterministic approximations are developed to solve the stochastic optimization problem over models described as arbitrary physics-based models where the optimization problem is to find the control setting that minimizes the expected cost subject to satisfying multiple deterministic and stochastic constraints with a given probability. The proposed algorithm is based on (1) a series of deterministic approximations to produce a candidate set of near-optimal control settings, and (2) stochastic simulations on the candidate set using optimal simulation budget allocation methods. Finally, the developed models and algorithms are used in a collaborative research effort to build a decision guidance system organized around a reusable repository of uniformly described models for composition, analysis, and optimization of manufacturing service networks.