Human in the Loop Generative AI for Production Rescheduling under Machine Disruptions

Human in the Loop Generative AI for Production Rescheduling under Machine Disruptions

Amirhossein Mirashrafi1

1) Master of Science (M.Sc.) in Industrial Management - Production and Operations Management, Allameh Tabatabaii University, Tehran

IOI Code : ACFZ-ACZKFE
Publication : The First Conference on Applied Research in Management, Economics and Humanities(carmeh.ir)
Abstract :
    Production rescheduling is required when an unexpected event, such as a machine breakdown, makes the current schedule no longer feasible. Many existing methods rely only on data recorded in the system, and generative AI, although it can propose new options, may also produce infeasible results. Shop-floor supervisors, meanwhile, have knowledge about repair duration, operator skill, and customer priority that the system does not record. This paper proposes a human-in-the-loop generative AI framework that divides decision responsibilities among three actors: generative AI proposes rescheduling options; a deterministic schedule checker verifies their feasibility, calculates their performance, and recommends one of the feasible options; and the supervisor adds tacit knowledge and makes the final decision. The framework is applied to an illustrative example of three machines and seven jobs, in which supervisor knowledge is added in four cumulative scenarios. In this example, the added knowledge changes the recommended schedule and halves its total weighted tardiness under the conditions known to the supervisor, while the checker rejects an infeasible proposal before it reaches the supervisor. One-factor experiments and a sensitivity analysis further show that the effect of each type of knowledge depends on the other conditions. These findings suggest that supervisor knowledge is context-dependent, that it is better added at the time of rescheduling, and that generative AI is better used to support supervisors than to replace them

Keywords : production rescheduling generative ai human in the loop machine breakdown tacit knowledge