A Robust Reactive Scheduling System with Application to Parallel Machine Scheduling

In this turbulent world, scheduling role has become crucial in most manufacturing production, and service systems. It allows the allocation of limited resources to activities with the objective of optimizing one performance measure or more. Resources may be machines in a factory, operating rooms in a hospital, or employees in a company, while activities can be jobs in a manufacturing plant, surgeries in a hospital, or paper work in a company. The goal of each schedule is to optimize some performance measures, which could be the minimization of the schedule makespan, the jobs\u27 completion times, jobs\u27 earliness and tardiness, among others. Until very recently, research has concentrated on scenarios that assume a predefined schedule that is failure free. Initial schedules produced in advance are being followed hoping no delays will occur, because once they do, the whole schedule may be compromised as it is not designed to adapt to change. Researchers focused on the generation of good schedules in the presence of complex constraints while assuming fixed processing times, known job arrival times, unbreakable machines, and immune employees. However, this is not the case in the real world, where processing times are stochastic, job arrival times could be unknown, machines do break down, and employees get sick. In fact, most environments including manufacturing are dynamic by nature and not static, vulnerable to many unpredictable events, which leads the initial schedule to become obsolete once it is executed. The reason these deterministic schedules fail is because they do not account for variability, scheduling the activities directly after each other, so when a certain activity is delayed, all its successors will be delayed too. In this dissertation, new repair and rescheduling algorithms, and robust systems equipped with learning capability are developed for the unrelated parallel machine environment, a known NP-hard problem. The introduced rules and algorithms were subjected to different stochastic rates of breakdowns and delays and were judged based on several performance measures to ensure the optimization of both the schedule quality and stability. Schedule quality is assessed based on the schedule Makespan (time to finish all jobs) and CPU, while schedule stability is based on the number of shifted jobs from one machine to another and the time to match up with the original schedule after the occurrence of a breakdown. The extensive computational tests and analyses show the superiority of the proposed algorithms and systems compared to existing methods in the literature, especially when implemented with the learning capability. Moreover, the rules were ranked based on their performance for different performance measure combinations, allowing the decision maker to easily determine the most appropriate repair/rescheduling rule depending on the performance measure(s) desired