Place of Origin
BS in Electrical Engineering
Sharif University of Technology
Entered program in 2013
Resource Allocation in Production Systems and Services under Uncertainty with Learning
Dissertation Summary: We study the problem of allocating resources to minimize costs or to maximize revenue in food-production industries and in services. We consider three different types of industries which face model uncertainty and require learning considerations. The first problem arises in distilleries and refineries, where a catalyst is used in a reactor to refine raw material. Catalyst performance decays as it is consumed. The productivity of a catalyst is not known a priori and is learned through production observations. We must decide when to replace the costly catalyst to minimize the total inventory costs and catalyst replacement costs. We develop a heuristic to solve this problem and a lower bound to evaluate the quality of the heuristic. We test our methods with real data from a leading food processing company and show our methods outperform current practice. The second problem is on production of products for which the value increases with age. Such products include whiskey and wine. We analyze the decision of a firm that is considering introducing an older aged product to the market. The older age product has uncertain demand and competes with the younger age product both in production capacity and in the market demand. The goal is to maximize the expected discounted revenue extracted from a fixed yearly production capacity, while considering uncertainty in demand, product substitution, and the learning process. We solve a simple version of the problem in closed form. We use the observations from the closed form solution, in addition to some theoretical properties to derive heuristics for more complex settings. The third problem is in cloud computing services. Customers require different classes of service. The service provider has a limited capacity per unit time and must allocate this capacity to arriving customers to maximize its revenue. There is uncertainty regarding completion times of each class of service, which requires learning over time.