Electrical Engineering Research Vol. 1 Iss. 4, October 2013 116 Fuzzy Inference System and Neuro‐Fuzzy Systems for Analog Fault Diagnosis Mohamed El‐Gamal *1 , Samah El‐Tantawy 2 *1,2 Engineering Mathematics and Physics Department, Cairo University, Faculty of Engineering *1 The American University in Cairo, *2 University of Toronto *1 Cairo, 2 Egypt *mhgamal@aucegypt.edu; 2 samah.el.tantawy@utoronto.ca Abstract A Fuzzy Inference System (FIS) is built to model and classify faults in analog circuits. The measurements that characterize the circuit under test (CUT) behavior are selected using feature extraction and dimensionality reduction techniques. These measurements are utilized to construct a rule based system that relates measurements (symptoms) to different faults (causes). In addition, hybrid neuro‐fuzzy systems are also constructed and trained to isolate the CUT faults. The integration of FIS and neural networks in these systems combines the remarkable pattern recognition capabilities of neural networks with the ability of fuzzy logic to incorporate and interpret linguistic knowledge. As a result, a superior diagnosis performance is obtained even if the CUT has overlapping faults. A benchmark circuit iS tested to demonstrate the high classification performance of the proposed procedure. Keywords Analog Circuits; Fault Diagnosis; Fuzzy Inference System; Neural Network; Neuro‐fuzzy System Introduction Fault diagnosis of analog circuits is conceptually divided into two main phases; fault detection and fault isolation. In the fault detection phase, the circuit is characterized as faulty. The faulty elements or regions are identified in the fault isolation phase. By a fault, we mean any change in the value of a circuit element with respect to its nominal value that can cause the failure of the circuit performance. Fault diagnosis of analog circuits is a challenging task due to the following reasons [Bandler & Salama (1985), El‐Gamal (1990)]:  The inaccuracy in circuit measurements besides the inability of measuring current without breaking the circuit connections.  The tolerance in circuit elements often complicates the fault diagnosis process.  The limited accessibility to circuit nodes especially in modern integrated circuits.  The lack of good fault models since analog circuits have a continuum of possible faults. Traditional methods are found inefficient in tackling the fault diagnosis problem. On the other hand, Artificial Intelligent (AI)‐based techniques are found promising in overcoming the difficulties stated above [Ahmed & Cheung (1994), El‐Gamal & Abu‐El‐Yazeed (1999)]. A comprehensive review of AI techniques used in fault diagnosis of analog systems is presented by Fenton et al. and Rutkowski and Grzecha [Fenton et al. (2001), Rutkowski & Grzecha (2008)]. Generally, AI techniques can be classified to traditional, model based, machine learning and soft computing‐based techniques. Fuzzy logic system, Neural Network (NN) and hybrid neuro‐fuzzy techniques are examples of soft computing‐based techniques. A fuzzy logic expert system is an expert system that uses fuzzy logic. In other words, a fuzzy expert system is a collection of fuzzy sets and rules that are used to reason about data. Fuzzy systems can be broadly categorized into two families. The first includes linguistic models based on collections of IF–THEN rules, whose antecedents (IF parts) and consequents (THEN parts) utilize fuzzy linguistic values (e.g. Mamdani FIS [Mamdani & Assilian (1975)]). The second category, based on Sugeno‐type systems [Takagi & Sugeno (1985)], uses a rule structure that has fuzzy antecedent and functional consequent parts. The analog faulty circuits are usually associated with imprecision and uncertainty. Therefore, the faulty circuit as is an excellent testbed for fuzzy systems [El‐ Gamal M.A., Abdulghafour (1996)]. Fuzzy logic systems behavior can be explained based on fuzzy