Foundation of Neuro-Fuzzy Systems and an Engineering Application

Abstract

This chapter discusses the foundation of neuro-fuzzy systems. First, we introduce Takagi, Sugeno, and Kang (TSK) fuzzy model [l,2] and its difference from the Mamdani model. Under the idea of TSK fuzzy model, we discuss a neuro-fuzzy system architecture: Adaptive Network-based Fuzzy Inference System (ANFIS) that is developed by Jang [3]. This model allows the fuzzy systems to learn the parameters adaptively. By using a hybrid learning algorithm, the ANFIS can construct an input-output mapping based on both human knowledge and numerical data. Finally, the ANFIS architecture is employed for an engineering example an IC fabrication time estimation. The result is compared with other different algorithms: Gauss-Newton-based Levenberg-Marquardt algorithm (GN algorithm), and backpropagation of neural network (BPNN) algorithm. Comparing these two methods, the ANFIS algorithm gives the most accurate prediction result at the expense of the highest computation cost. Besides, the adaptation of fuzzy inference system provides more physical insights for engineers to understand the relationship between the parameters.