Fitting an uncertain productivity learning process using an artificial neural network approach

Abstract

Productivity is critical to the long-term competitiveness of factories. Therefore, the future productivity of factories must be estimated and enhanced. However, this is a challenging task because productivity can be improved based on a learning process that is highly uncertain. To address this problem, most existing methods fit fuzzy productivity learning processes and convert them into mathematical programming problems. However, such methods have several drawbacks, including the absence of feasible solutions, difficulty in determining a global optimum, and homogeneity in the solutions. In this study, to overcome these drawbacks, a specially designed artificial neural network (ANN) was constructed for fitting an uncertain productivity learning process. The proposed methodology was applied to an actual case of a dynamic random access memory factory. Experimental results showed that the ANN approach has a considerably higher forecasting accuracy compared with several existing methods.