Fast Generation of Optimal Speed Command of a Cascaded Conveyer Belt System based on Linear Programming and Deep Learning

Abstract

In this paper, we construct a simple deep learning model for quickly transporting objects that arrive at irregular times, and for adjusting them at a certain period (or it's positive integer multiple) by using n conveyor belts arranged in a row. Previously, we proposed a method for determining the speed of each conveyor belt by iterative calculation of Linear Programming, but there was a trade-off between optimality and the amount of calculation. Here, we treat the Linear Programming based algorithm as an input-output system and show that it can be replaced by the simple deep learning model with sufficient accuracy. As a result, we show that the speed command value for one object can be generated in less than about 1/1000 of the time required by the conventional method, and that throughput is improved.