Smart Factory Performance: Data-Driven Scheduling and Real-Time Optimization

Thesis Abstract

The rapid evolution of manufacturing ecosystems toward interconnected machines, real-time data streams, and autonomous decision-making presents a critical challenge achieving synchronized production schedules that adapt dynamically to variability while maintaining quality, throughput, and energy efficiency. Despite advances in Industry 4.0, many factories struggle with static master schedules that fail to exploit real-time information, resulting in suboptimal utilization of resources, increased lead times, and higher operational costs. This study addresses the gap by combining data-driven scheduling with real-time optimization to enhance smart factory performance through a unified framework that integrates shop-floor telemetry, demand signals, and energy metrics. The aim is to design, implement, and evaluate an integrated scheduling and optimization framework that leverages streaming data for dynamic rescheduling and real-time resource allocation. Specific objectives are (i) to develop a data fusion architecture that ingests shop-floor sensors, MES/ERP events, and external demand signals; (ii) to formulate a multi-objective scheduling model balancing throughput, due-date adherence, and energy consumption; (iii) to implement real-time optimization using a rolling-horizon approach with model predictive control and a lightweight metaheuristic for rapid decision-making; (iv) to validate the framework in a pilot-scale manufacturing cell comprising 12 machines and 6 work-centers with heterogeneous processing times; (v) to compare the performance against a baseline static schedule using simulated and live production data; and (vi) to assess operational robustness under disturbances such as machine failures and rush orders. The methodology adopts a mixed-methods research design anchored in a pragmatic paradigm. The population includes a medium-sized discrete-manufacturing facility implementing an automated production line. A purposive sample of live production data from 12 months prior to and 6 months post-implementation will be used, comprising approximately 1.2 million event logs, 3,500 work-orders, and energy consumption records at 1-second granularity. Data collection instruments consist of (a) historians and MES interfaces for real-time data streams, (b) structured interviews with shop-floor supervisors to capture tacit knowledge on constraint handling and responsiveness, and (c) process documentation for validation of model assumptions. Validity and reliability are ensured through cross-validation of event timestamps, calibration of machine models with observed cycle times, and test-retest checks of scheduling outputs under simulated disturbances. Analytical techniques include (i) descriptive analytics to characterize baseline performance and variability; (ii) regression analysis to quantify relationships between scheduling decisions, throughput, and energy metrics; (iii) time-series analysis to detect trends and seasonality in demand and utilization; (iv) multi-objective optimization using a hybrid approach that combines model predictive control (MPC) with a genetic algorithm to generate near-optimal schedules within real-time constraints; (v) Monte Carlo simulations to evaluate robustness under stochastic disturbances; and (vi) ANOVA and paired t-tests to assess significance of improvements over the baseline. A data fusion framework based on an event-graph representation and a streaming ETL pipeline will be implemented to support real-time decision-making. The conceptual framework integrates the Theory of Constraints (TOC) for constraint-aware scheduling and the Resource-Based View (RBV) of firm capability, augmented by the Dynamic Capabilities Theory to justify organizational adaptation to digitalized operations. Expected findings include demonstrable improvements in on-time delivery, throughput, and energy efficiency, with target gains of 12–18% in throughput, 8–15% reduction in energy consumption per unit, and a 20–30% reduction in due-date violations under typical disturbances. The research is anticipated to reveal that the proposed rolling-horizon MPC with adaptive priority rules outperforms static schedules by maintaining higher machine utilization and shorter makespans while limiting re-entrant work and changeover costs. The study will contribute to knowledge by providing a rigorous, data-driven blueprint for integrating real-time optimization with scheduling in smart factories, articulating a practical methodology for practitioners and expanding the empirical literature on digital twins, intelligent scheduling, and energy-aware manufacturing. The main conclusion is that a tightly coupled data-driven scheduling and real-time optimization framework significantly enhances factory performance in dynamic environments, given reliable data governance and timely feedback loops. Recommendations include implementing scalable data architectures, investing in sensor fidelity and data quality, developing organizational protocols for real-time decision authority, and extending the framework to multi-site operations with transfer-learning capabilities. Suggestions for future work include exploring reinforcement learning-based schedulers, incorporating human-in-the-loop interfaces, and evaluating long-term total-cost-of-ownership impacts.