SOURCE:

Faculty: Engineering
Department: Electronic And Computer Engineering

CONTRIBUTORS:

Chiagunye, T. T.
Inyiama, H. C.

ABSTRACT:

This research is concerned with the modeling of agent-based job shop scheduling and control using Markov chain for steady state probabilities. The model agent-based job shop scheduling involves three sequential machines through which every order must pass followed by one out of three finishing machines used one per finishing type. It was mandatory for the type of aluminum sheets being produced that the raw material be passed through the first three machines only in one order. Thus the model developed took this sequential order into consideration. This model was executed using a combination of Markovian process, in working out the state of the machine and agent-oriented analysis that adjusts to the dynamics of the stochastic order processes. The steady state probabilities of the machines as worked out by Markovian process were found to be (A) 0.0408, (B) 0.0750, (C) 0.1794, (D) 0.6983 and (E) 0.0065. The values represent the depreciation rates of the machines in use which help to determine the real cost of each order processed. The extra raw material needed to augment for wastages in a production plant worked out by Markovian process was found to be 11 kg extra per 100kg of output. A well-crafted scheduler agent carries out bunching of sorted jobs either in 1 or 2 or 3 days’ bunch(es) per finishing type and selects the best out of the three approaches. This scheduling technique allows a certain product type to be scheduled for 1 or 2 or 3 days before changing to another product type. The result of ten different monthly orders scheduled with bunching factor 2 had earliest release dates for eight out of the ten different orders and bunching factor 3 had earliest release dates for two orders while bunching factor 1 had none. Because the simulation is done before the production, the scheduler works out the best bunching factor for any given order and recommends that for use in the scheduling. The agent-based job shop scheduling model was validated with D.G. Kendall, classical method for poisson arbitrary distribution with nonpreemptive discipline where the agent-based model (ABM) compared favorably with the classical model. ABM had better results in orders 4,6, 8, 9, 10 and approximately equal to the classical method in orders 1, and 3. The classical method is only better in orders 2, 5 and 7. The comparative result shows that the modelled agent-based job shop scheduling had 2.4% improvement to the existing classical model and should be applied in an industrial set up for optimum machine usage and customer satisfaction. The simulation results were also used to determine the optimum scale of plant, given the rate of order arrival per month.