Generalized job-shop scheduler using feed forward neural network and greedy alignment procedure

This paper describes a generalized job shop scheduler using a 3-layer feed-forward Neural Network (NN) and a greedy alignment procedure. The NN is used to detect precedence order of operations within each job which is humanly subjective in nature. The greedy alignment procedure aligns operations of various jobs on respective machines with feasible/optimal Finishing Time (FT). The greedy alignment procedure is designed in such a way that it tries all possible combinations of operation alignments on given machines and estimates FTs in order to confirm whether a shop has more than one most feasible or optimal schedule or not. We propose two types of scheduling in this paper: Schedule all jobs at a time on given machines and estimate the FT, or schedule the shop by setting all machines with the same operation duration by sharing sections of waiting operations to free predecessor machines and estimate the FT. Simulation results show that sharing scheme is the best one for feasible/optimal schedules.