Robust hybrid architecture to signals from manufacturing and machine monitoring sensors

Article type: Research Article

Authors: Hambaba, Ahmed

Affiliations: National Research Council (NRC), NASA Ames Research Center, Ames, CA, USA, and Computer, Information, and Systems Engineering Department, San José State University, San José, CA 95192-0180, USA. E-mail: hambaba@sjsu.edu

Abstract: The design and implementation of effective control of manufacturing processes depends on the successful monitoring and recognition of process signals. In this paper, we describe an application of robust hybrid model to the prediction of plasma etcher outcome; etcher rate, selectivity, and uniformity. The robust hybrid model consists of two main architectures: robust model and recurrent neural network with real-time recurrent learning. The residual errors are used for the tuning of the weights of the recurrent neural network. The robust hybrid model's performance is compared to Ordinary Least-Square method, Alpha-trimmed mean, and Back-Propagation neural network.

Journal: Journal of Intelligent and Fuzzy Systems, vol. 9, no. 1-2, pp. 29-41, 2000