Fault diagnosis of rolling bearings using a genetic algorithm optimized neural network

• Published in: Measurement : journal of the International Measurement Confederation Vol. 58; pp. 187 - 196
• Main Authors: Unal, Muhammet, Onat, Mustafa, Demetgul, Mustafa, Kucuk, Haluk
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2014
• Subjects: Envelope detector; Fast Fourier Transform artificial neural network; Fault diagnostics; Feed-forward algorithm; Genetic algorithm; Hilbert Transform; Rolling bearing; Envelope detector; Feed-forward algorithm; Hilbert Transform; Genetic algorithm; Fast Fourier Transform artificial neural network; Fault diagnostics; Rolling bearing
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2014.08.041

[Display omitted] •Fault diagnosis system using feature extraction and classification techniques.•Vibration signals were analyzed by using the Hilbert Transform.•Features of vibration signals were extracted by FFT analysis.•Trial and error time is significantly reduced by GA–ANN with the high accuracy.•GA–ANN classification algorithm performance is tested in the various experiments. In rotary complex machines, collapse of a component may inexplicably occur usually accompanied by a noise or a disturbance emanating from other sources. Rolling bearings constitute a vital part in many rotational machines and the vibration generated by a faulty bearing easily affects the neighboring components. Continuous monitoring, fault diagnosis and predictive maintenance, is a crucial task to reduce the degree of damage and stopping time for a rotating machine. Analysis of fault-related vibration signal is a usual method for accurate diagnosis. Among the resonant demodulation techniques, a well-known resolution often used for fault diagnosis is envelope analysis. But, usually this method may not be adequate enough to indicate satisfactory results. It may require some auxiliary additional techniques. This study suggests some methods to extract features using envelope analysis accompanied by Hilbert Transform and Fast Fourier Transform. The proposed artificial neural network (ANN) based fault estimation algorithm was verified with experimental tests and promising results. Every test was initiated with a reference ANN architecture to avoid inappropriate classification during the evaluation of fitness value. Later, ANN model was modified using a genetic algorithm providing, an optimal skillful fast-reacting network architecture with improved classification results.