Date on Master's Thesis/Doctoral Dissertation

5-2025

Document Type

Doctoral Dissertation

Degree Name

Ph. D.

Department

Industrial Engineering

Degree Program

Industrial Engineering, PhD

Committee Chair

Parikh, Pratik J.

Committee Co-Chair (if applicable)

Aqlan, Faisal

Committee Member

Segura, Luis J.

Committee Member

Noor E Alam, Md

Author's Keywords

Dynamic assignment; Dual bin packing; lookahead; bi-objective optimization; energy-efficient scheduling; lexicographic optimization

Abstract

Scheduling in online and dynamic manufacturing systems presents significant challenges due to the continuous arrival of customer orders with certainty only over a short time window and the dynamic availability of resources after completing ongoing operations. Similar challenges exist at our industry partner’s site when scheduling orders (computer servers) to machines (testing cells) during the final testing process before customer delivery. The primary challenge lies in dynamically assigning incoming orders to available machines in order to minimize the number of unfinished orders. Additionally, the energy-intensive nature of the production process introduces another challenge; efforts to reduce energy consumption often lead to increased tardiness, creating a trade-off between scheduling efficiency and sustainability goals. Despite these challenges, existing scheduling models struggle to effectively incorporate short-term order certainty and its variations while addressing key real-world system and operational constraints, such as skewed order arrival patterns, order-to-machine compatibility, and physical order allocation. Furthermore, there are no known methods to appropriately trade-off energy consumption and tardiness in online and dynamic settings, while accounting for conditional and machine-specific energy consumption, as well as dynamic processing times driven by real-time order-to-machine assignments. To address these gaps, this dissertation develops optimization-based approaches that integrate lookahead and machine learning-driven methodologies, making two key contributions to both academic literature and industry practice, as outlined below. Our first contribution (Chapter 2) introduces an Online Dynamic Dual Bin Packing with Lookahead (OD-DBP-LA) framework, a generalization of the Online Bin Packing with Lookahead (OBP-LA) approach, to optimize order-to-machine assignment along three directions: (i) time lookahead, (ii) dual bin packing (DBP), and (iii) the dynamic concept (D). We propose an integer linear programming model that incorporates key real-world system and operational constraints, including skewed order arrival patterns, order-to-machine compatibility, variable machine capacities, physical allocation of orders, and multiple order types. To efficiently solve each deterministic, lookahead window-specific subproblem, we develop a ‘2-phase’ computational framework that integrates mathematical programming and a metaheuristic, Genetic Algorithm. Our results show that the length of the lookahead window, testing capacity, order arrival patterns, test processing time requirements, and physical allocation of orders significantly impact scheduling performance. Our second contribution (Chapter 3) develops a bi-objective optimization model to minimize tardiness and energy consumption. Unlike traditional approaches, our model dynamically adjusts processing times and energy consumption based on assignment decisions. To prioritize the timely fulfillment of customer orders over energy efficiency, we employ a lexicographic multi-objective optimization approach and a lookahead strategy to handle the real-time uncertainty of continuous arrivals, sequentially minimizing total tardiness followed by total energy consumption within each lookahead window. To efficiently solve industry-scale problems, we propose a Machine Learning-Matheuristic (ML-Matheuristic) approach integrating mathematical programming, Evolutionary Particle Swarm Optimization (EPSO), and machine learning. Additionally, our proposed EDD-based dynamic machine learning (EDD-DML) dispatching rule, achieves over 90% reduction in computational time, with only ~10% increase in tardiness and ~2% in energy consumption. Our findings contribute to the online and dynamic manufacturing optimization literature by developing scalable, data-driven decision-support approaches that enhance real-time scheduling efficiency and help determine the optimal decision-making frequency to improve outcomes. Furthermore, decision-makers can weigh the trade-off between investing in information technology to increase order arrival certainty and expanding system capacity to accommodate smaller demand sizes. In systems with excess capacity, decision-makers can significantly reduce energy consumption without increasing tardiness, which is particularly valuable for industries managing seasonal demand variations or scaling up capacity to enhance system adaptability. Finally, proposed approaches also enable decision-makers to negotiate delivery windows or due date penalties with clients, ensuring customer satisfaction while achieving energy goals.

Download

Included in

Engineering Commons

Share

COinS