Data mining based on statistical parameters to improve fault diagnosis accuracy

Vibration signals contain rich information about the health of machinery. There for vibration condition monitoring is used in industries. In present study vibration signals from gearbox of Massey Ferguson 285 tractor is gained in three health condition of a gear: Healthy, Worn tooth face and Broken tooth. Vibration signals are turned to frequency domain by applying a Fast Fourier Transform (FFT) to them. Then some statistical parameter is used for data mining from the signals. Processed signals are used as input vectors for Feed Forward Back-propagation neural networks with variable hidden layer neurons count between 1 and 10 in 2 main structures, two and three layers network. Maximum 100% classification accuracy gained from two-layer network with 4 hidden layer neurons and three-layer network with 3x3 and 8x7 hidden layer neurons.

Keywords:

Vibration Condition Monitoring , Fault classification , Neural Network , Data mining