Using Parameter Prediction Models to Forecast Post-Interruption Learning

Abstract

Industrial learning curves are well-established tools for estimation of labor times and costs during the start-up, or learning, phase of a production activity. Their accuracy is impressive given a stable environment and continuous production. One of the problems that users face, however, is dealing with interruptions of the learning process. An improved ability to predict the course of learning after interruptions of production would enable manufacturers, service companies and others to improve their cost estimates and plan for resource requirements. Recent research has addressed the effects of interruption (and forgetting) as well as the nature of the relearning curve. This project represents an attempt to predict the parameters of relearning curves based on information available at the time that production resumes—including the original learning curve parameters, the amount of original learning, and the length of the production interruption. We develop and test parameter prediction models (PPMs) for estimating the parameters of relearning curves, then compare the predictive ability of such “estimated relearning curves” to the predictive ability of relearning curves (RLCs) that use only the relearning data. The PPMs perform well, showing lower mean absolute percentage errors than the best RLC model for this task until several relearning data points become available.