Multilayer fuzzy neural network for modeling a multisource uncertain unit-cost learning process in wafer fabrication

Tin-Chih Chen, Horng Ren Tsai*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Purpose: The purpose of this study is to model a multisource uncertain unit-cost learning process to estimate the future unit cost of manufactured products. Design/methodology/approach: A multilayer fuzzy neural network (FNN) is constructed to model a multisource uncertain unit-cost learning process. A fuzzy constrained gradient descent algorithm is proposed to train the FNN. Findings: The proposed methodology was applied to a wafer fabrication factory. Wafer fabrication, a well-known additive manufacturing process, is a highly competitive industry; therefore, the manager of a wafer fabrication factory is concerned about the unit cost of each product. This cost can be reduced through learning processes, but these involve much uncertainty, making the estimation of the unit cost a challenging task. Existing methods for modeling these processes and outcomes cannot account for multiple learning sources. However, the multilayer FNN constructed in this study successfully addressed these problems and improved the accuracy of the unit cost estimation by 88 per cent in a real case study. Originality/value: Modeling an uncertain unit-cost learning process is an innovative application of an FNN. In addition, the proposed methodology is the first attempt to separate the effects of several learning sources, which is considered conducive to the estimation performance.

Original languageEnglish
Pages (from-to)521-531
Number of pages11
JournalRapid Prototyping Journal
Volume24
Issue number3
DOIs
StatePublished - 1 Jan 2018

Keywords

  • Estimation
  • Fuzzy neural network
  • Learning model
  • Unit cost

Fingerprint Dive into the research topics of 'Multilayer fuzzy neural network for modeling a multisource uncertain unit-cost learning process in wafer fabrication'. Together they form a unique fingerprint.

Cite this