A fuzzy syntactic approach to fault diagnostics by analysis of time-sampled signals
Abstract
This work presents the FSDS, a diagnostic tool which follows a fuzzy syntactic approach to fault diagnostics by analysis of time-sampled signals. The FSDS uses the signals generated by the target system to achieve the diagnosis. First, the FSDS transforms the signal into a string of sets of elementary structures, templates. Then, examining the consecutive templates, the FSDS detects whether or not the string characterizes a certain condition. After the syntactic analysis is performed for each desired condition, the FSDS lists the possible conditions of the system under analysis along with the attached possibilities. The FSDS consists of two components: the lexical analyzer and the syntactic recognizer. The lexical analyzer transforms the signal into a string of template sets. Each set of templates in the string represents a certain window of samples of the original signal. Each template in a set is assigned a weight by the lexical analyzer to show the strength of the match of the template to the corresponding window of the original signal. Since each window in the original string is represented by multiple templates, the string is termed multi-category string. Each alphabet template corresponds to an ART2 category. When the input time-sampled signal is transformed into the multi-category string, this string is presented to the syntactic recognizer of the FSDS. The syntactic recognizer analyzes the multi-category string to see if the string characterizes a condition. After the analysis is repeated for each desired condition, the recognizer lists the possible conditions that exist in the target system currently. The recognizer achieves the syntactic analysis by a set of hierarchical fuzzy state machines (FSMs). Each condition has a set of FSMs specifically designed to recognize the strings that characterize this condition. FSMs are built manually using statistics collected from the results of the lexical analysis. Two applications to ECG diagnosis and SRM phase current analysis are also presented. The results have shown that the FSDS was able to distinguish the atrial fibrillation from the normal sinus rhythm. Furthermore, the FSDS was able to distinguish normal phase current from one with partial phase short.
This paper has been withdrawn.