COBEM99

SR21 Redes Neurais, Algoritmos Genéticos e Lógicas Nebulosa/Neural

Title:

APLICATION OF NEURAL NETWORK IN ENGINE FAULT DETECTION

Summary :

ABSTRACT. THIS WORK PRESENTS THE IMPLEMENTATION OF AN AUTOMATIC DIAGNOSIS TOOL FOR SOME FAULTS FOUNDED IN SPARK ENGINE. USING THE MAGNITUDE OF ENVELOPE SPECTRUM, IN CHARACTERISTIC FREQUENCIES, CERTAIN PATTERNS ASSOCIATED TO THE DEFECTS MORE FREQUENTLY FOUND DURING THE TESTS IN THE ASSEMBLY LINE WERE DEFINED. GROUPS OF THESE PATTERNS WERE PRESENTED TO A NEURAL NETWORK OBJECTIFYING ITS TRAINING. THE TYPE OF NEURAL NETWORK ADOPTED, WAS THE PROBABILISTIC NEURAL NETWORK(PNN), WHICH PRESENTS DESIRABLE CHARACTERISTICS TO SOLVE THE PROBLEM IN FOCUS: THE RELIABILITY OF THE STATISTICAL MODEL IN WHICH IT IS BASED, THE ADDITION OR REMOVAL EASINESS OF PATTERNS IN DATA SET OF TRAINING AND VALIDATION, AND THE REDUCED COMPUTATIONAL TIME EXPENDED IN ITS TRAINING. THE WORK FINISHES SHOWING A VALIDATION PROCEDURE, IN WHICH THE CAPACITY OF THE TRAINED NET IN RECOGNIZING PATTERNS THAT ARE DIFFERENT FROM THOSE USED IN ITS TRAINING, WAS TESTED. THE PNN WAS CAPABLE TO AUTOMATE THE FAULT DIAGNOSIS, ESTABLISHING, LIKE THIS, AN OBJECTIVE APPROACH FOR THE CONTROL QUALITY OF SPARK ENGINES CONDITIONS. KEY WORDS: AUTOMATIC FALT DETECTION, NEURAL NET WORKS, SPARK ENGINE

Author :

Mancuzo, Mechelangelo Viana
Ribeiro, Carlos Roberto

Paper View :