Fault diagnosis using Neuro-fuzzy Transductive Inference Algorithm

The primary goal of this research is to develop a novel intelligent fault diagnosis method employing Neuro-fuzzy Transductive Inference algorithm (NFTI) in order to solve the the global model application problem, as well as the global availability of the model and sample data set. The method is characterized by that a personal local model which is established for every new fault symptom input data in the fault diagnosis systems, based on some closest samples next to this fault symptom data in an existing sample database. Compared with other similar inductive method (ANFIS - Adaptive Neuro-Fuzzy Inference System) on Fisher's Iris data set, the mentioned algorithm classifier has reduced 15% of the average test error and increased approximately 30% of classification speed. Detecting the fault symptom data set sampled from actual aeronautic thrustor test, the presented system can identify accurately three fault states. The results of the research indicate that the availability and efficacy of the fault diagnostic strategy is superior to any other inductive reasoning technique about some fault diagnosis issues.