Analysis and scheduling of intelligent manufacturing systems

Abstract

This thesis addresses the analysis and scheduling challenges in intelligent manufacturing systems, focusing on three critical characteristics: delayed differentiation, production-inventory interactions, and flexible workforce scheduling. By integrating theoretical modeling, dynamic behavior analysis, and optimization frameworks, this work advances the understanding and practical application of intelligent manufacturing systems under real-world constraints.

First, we propose a fast algorithm for predicting the production process performance in the delayed differentiation-based Flexible Manufacturing System (FMS) under operation control. We formulate practical problems existing in the auto, food, and furniture industries into a mathematical formation. Then, we offer closed-form formulae for predicting the production process performance using the built stochastic model in the production line with three machines. We also propose an algorithm to predict the performance of a production line having more than three machines. The proposed methods were verified to be highly accurate through comparison experiments.

Second, we develop effective algorithms for analyzing the dynamic behavior in manufacturing systems with regular orders. In addition to machines and work-in-process buffers usually considered in current literature, a Finished Goods Buffer (FGB) is added to the manufacturing system to store the final products. For a fixed shipping period, a certain amount of products are shipped to the customers. The system with one machine and finished goods buffer isolated from the manufacturing system is first modeled and estimated. Then, based on it, we proposed an aggregation method, which can be employed in larger systems. The approach substantially decreases the state space of the problem and leads to the modeling and estimation of the system solvable. Numerical experiments show that the proposed methods maintain high accuracy.

Third, we study state modeling and dynamic analysis in production-inventory systems with geometric machines. Specifically, we study the production-inventory system with only one machine to offer a building block for larger lines. Then, an approximation method is proposed to model and analyze the system with more than one machine by using the ideas of decomposition and aggregation. The developed approach substantially solves the problem of state explosion, leading to this problem being solvable. Simulation experiments show that the proposed algorithm is highly accurate.

Finally, we investigate the optimal scheduling policies for a flexible worker in the two-station manufacturing system with finite buffers and setup time. We describe the practical observation in a truck production line in mathematical form. Then, we identify the optimal form and parameter for a system. We find that static scheduling is typically optimal when reassignments are costly, while dynamic scheduling is generally optimal when reassignments are low cost. The flexible worker can be optimally scheduled according to the work-in-process level in the buffer (when dynamic scheduling is optimal form) or the system's parameters (when static scheduling is optimal form).