A fuzzy back propagation network for output time prediction in a wafer fab

摘要: Abstract Output time prediction is a critical task to wafer fab. Three major techniques commonly applied predict the output of lot include multiple-factor linear regression, production simulation, and back propagation networks (BPNs). Among them, network has been shown be most promising approach for practical applications, considering both accuracy efficiency. In this study, fuzzy (FBPN) constructed incorporate production-control expert judgments in enhancing performance an existing crisp network. Parameters chosen as inputs FBPN are no longer considered equal importance, but some experts requested express their opinions about importance each input parameter predicting with linguistic terms, which can converted into pre-specified numbers, aggregated, then multiplied normalized value corresponding when fed FBPN. Subsequently, arithmetic triangular numbers deal all calculations involved learning. A defuzzification operation finally performed obtain forecast enhance practicability forecast. fab simulator generate test examples. The results five cases showed outperformed BPN efficiency respect, including starting much smaller initial root mean square error (RMSE) requiring fewer epochs convergence. respect measured minimal RMSE, was slightly better than that BPN. addition, our experiments also it possible significantly reduce RMSE if fuzzy-valued treated weighted interval actual cycle time.