Forecasting learning curves
It is well known that performance improves with practice. In manufacturing activities this phenomenon is variously described as a learning curve, progress function, or experience curve, depending on the operational circumstances and viewpoint of the observer. This paper is concerned with forecasting future performance in such situations. It is an important industrial requirement for many purposes, including costing, process capacity planning, manpower planning, batch sizing and delivery date projections. The approach adopted is to use a learning curve model in Industrial Dynamics format. For most practical studies modelled by the present author, the Time Constant Model has been found to describe start-up performance adequately. The various parameter estimation algorithms described in the paper concentrate on this model, although the procedures adopted are of much wider application. It must be emphasised that the problem may be expressed as parameter estimation during the early part of the transient response in circumstances where the signal-to-noise ratio may be relatively poor, i.e., the detection of rapidly changing trends in the presence of considerable scatter.