Learning curve parameter estimation beyond traditional statistics

Due to decreasing order quantities, increasing product variety and fluctuating production orders, manufacturing companies have been encountering an increased occurrence of repetitive learning-forgetting phenomenon. In this paper, deterministic methods for the learning curve parameter estimation from the limited production data available from the unstable production environment are studied. Two main learning curve models: cumulative average (Wright) and unit (Crawford) were considered and several different mathematically proven methods were proposed for the parameter estimation. The calculation results illustrated that learning curve parameters can be unequivocally estimated from the limited production data (single random sample) by using deterministic methods for both of the learning curve models, although more accurate estimation was provided by the cumulative average model based methods. Newly proposed methods enable sufficiently accurate parameter estimation from the limited production data where traditional statistical parameter estimation methods cannot be applied.