Estimating the production time of a PCB assembly job without solving the optimised machine control

Abstract

Production planning and control of the printed circuit board (PCB) assembly includes several decisions dealing with, for example, grouping of PCB jobs, allocation of PCB batches to machine lines, sequencing of batches and load balancing of lines. The production time of a PCB job for a given placement machine is a key factor in this context and it must be quickly and accurately estimated, possibly millions of times in a single planning task, to avoid erroneous decisions. The commonly used nominal tact time-based estimators are very rough and the machine simulators too slow. Therefore, the purpose of this study is to give better machine-specific estimators that avoid the construction the actual machine control program. Two new estimators are proposed for gantry machines, one based on the information given by the manufacturer about the operations of the placement head, and the other on the regularised least-squares regression method trained with a set of PCB placement jobs. In practical evaluation with 95 PCB jobs, the mean absolute percentage error of the first and second methods are 3.75% and 6.52%, respectively, while that of the tact time-based approach is more than 17%. This indicates a great potential of the proposed methods as production time estimators.